AI-driven approaches for dysgraphia diagnosis using online and offline handwriting data: A comprehensive scoping review Authors: Avisa Fallah, Yazdan ZandiyeVakili, Hedieh Sajedi, Sadiq Abdulhussain, Sadiq Abdulhussain, Sadiq Abdulhussain, Sadiq Abdulhussain, Sadiq Abdulhussain Published: 2025 ABSTRACT Dysgraphia, a neurodevelopmental disorder impacting writing abilities, is often overlooked or misdiagnosed, affecting academic and daily life activities. It affects both children, especially school-aged, and adults, potentially stemming from brain trauma or neurological diseases. Early diagnosis is crucial for effective management and intervention. With advancements in artificial intelligence (AI), there is potential to improve early diagnosis and intervention through automated detection methods. This scoping review aims to explore available studies employing AI-based models for dysgraphia diagnosis and identify the highest performance models and their challenges to suggest future improvements. Comprehensive searches were conducted across PubMed, Scopus, Web of Science, IEEE Xplore, and SpringerLink until April 2024. Included studies are original research articles focusing on dysgraphia detection using AI-based predictive models with image processing techniques on handwritten images. Data extraction was done using a structured form in Google Sheets, capturing study characteristics, datasets, methods, and results. Out of 177 initial papers, 21 met the inclusion criteria. Studies span nearly a decade, with 38% being conference papers and 62% journal articles from various countries. AI models, particularly Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs), demonstrated high accuracy rates, often surpassing 90%, with comprehensive feature extraction methods enhancing performance. Significant challenges include small sample sizes and language-specific models. AI-based models, especially advanced ones like CNNs and SVMs, significantly enhance dysgraphia diagnosis, offering faster and more accurate assessments than traditional methods. Future research should focus on larger, more diverse datasets, language-independent models, and standardized evaluation metrics. Integrating AI-based tools in educational and healthcare settings can revolutionize dysgraphia management, improving academic outcomes and support for affected individuals. 1. Introduction Writing, a crucial skill developed early in life, is linked to overall achievement, particularly in academic pursuits, and can take up to fifty percent of the school day [1,2]. Poor handwriting can lead to lower self-esteem and social difficulties, with students frequently misunderstood as careless or unmotivated instead of identified as having a learning disability [3–5]. Dysgraphia (Fig 1) is a neurodevelopmental disorder predominantly affecting boys, impairing writing abilities from childhood through adulthood [6,7]. It may arise from brain trauma, infections, or diseases that harm the parietal lobe. It is commonly observed in conditions like cerebrovascular injury, Huntington’s disease, multiple sclerosis, Alzheimer’s disease, vascular dementia, and Parkinson’s disease [8]. Symptoms include problems with letter formation, legibility, spacing, spelling, and grammar that do not align with the individual’s cognitive levels despite appropriate learning [7,8]. Sample handwritten letter from a patient diagnosed with dysgraphia [19]. Recent research suggests that various subtypes of dysgraphia may be linked to distinct neurological mechanisms [9], with significant interrelations among brain regions governing routine processes, linguistic abilities, and movement coordination, particularly the cerebellum [8]. Functional imaging and case studies indicate that cerebellar damage can lead to acquired dysgraphia symptoms, underscoring its role in language and automaticity, especially in coordinating writing [10,11]. The cerebellum may help develop a neural system critical for writing, disruption of which could cause various functional impairments [12]. Additionally, genetic studies highlight a hereditary component in verbal executive functions, orthographic skills, and spelling abilities [13–15]. Notably, genes on chromosomes 15 and 6 have been linked to difficulties in reading, spelling, and phonemic awareness, respectively [16,17]. Often misdiagnosed or overlooked, dysgraphia frequently coexists with other learning and psychiatric disorders and can have lifelong impacts and hinder vocational progress and daily activities in adults [18]. Effective management of dysgraphia, crucially involving the educational system, is essential for proper diagnosis and treatment. Dysgraphia diagnosis requires a team effort from specialists in education, psychology, and medicine to evaluate a student’s handwriting abilities and discern contributing factors to their writing challenges [20]. Before a thorough evaluation, it is crucial to rule out other conditions that could affect handwriting, such as hearing loss, visual impairments, or inadequate training [21]. Key assessment areas include writing speed, clarity, spelling consistency, and pencil grip to diagnose dysgraphia effectively [22,23]. Although a universally standardized medical assessment method for diagnosing dysgraphia does not yet exist, recent advances in artificial intelligence suggest that machine learning (ML) could be a viable solution for early detection [5,22,24]. These methods could offer faster and more accurate evaluations than traditional assessments by experts, which are subject to human biases and the variability of specialist expertise. Additionally, manual assessments demand considerable time and human resources. In our scoping review, we aim to identify and explore valuable research that employs computational approaches through AI-based models for dysgraphia detection in handwriting images. As this area of research is relatively unexplored and lacks in-depth discussion, this review aims to identify available studies, collect related information, especially the datasets used, crucial features, and challenges of building these models, and employ a variety of methods and study designs to address broad and complex questions, rather than focusing on specific outcomes. We will also discuss future trends in dysgraphia detection models based on handwriting and other methods for recognizing dysgraphia beyond handwriting images. Thus, we conducted this scoping review instead of a systematic review. 1.1. Objectives 1Identify available studies that used AI-based models for dysgraphia diagnosis and explore the current dominant use of AI in expediting dysgraphia detection. Identify available studies that used AI-based models for dysgraphia diagnosis and explore the current dominant use of AI in expediting dysgraphia detection. 2Explore the challenges of building computational AI-based models under current conditions and the limitations of each study. Explore the challenges of building computational AI-based models under current conditions and the limitations of each study. 3Determine the best models with high accuracy and performance in dysgraphia diagnosis, examine their effectiveness in dysgraphia detection, and consider possible future improvements in these models. Determine the best models with high accuracy and performance in dysgraphia diagnosis, examine their effectiveness in dysgraphia detection, and consider possible future improvements in these models. 4Examine the integration of machine learning techniques with traditional diagnostic methods. Examine the integration of machine learning techniques with traditional diagnostic methods. 5Analyze the training datasets used for developing AI models in dysgraphia diagnosis. Analyze the training datasets used for developing AI models in dysgraphia diagnosis. 6Review technological advancements and their implications for future AI models in dysgraphia diagnosis. Review technological advancements and their implications for future AI models in dysgraphia diagnosis. 2. Methods The primary aim of this scoping review is to investigate, identify, and evaluate mainstream AI-based models used to detect dysgraphia through the analysis of handwritten images. Based on our aim, a scoping review focusing on leading and groundbreaking models that have emerged over the past decade was undertaken, aligning with the guidelines reported in the Scoping Reviews (PRISMA-ScR) Statement [25]. 2.1. Eligibility criteria All original research articles published until April 2024 that explore dysgraphia detection using predictive AI-based models with image processing techniques on handwritten images were included. This criterion ensures a focused review of advanced AI methodologies for diagnosing dysgraphia. Papers were excluded if they did not use AI-based predictive models for dysgraphia detection or relied on other data sources like EEG or brain scans. Additionally, review articles, studies on other learning disabilities (e.g., dyslexia, dyscalculia), gray literature, and conference abstracts were excluded to maintain a clear focus. 2.2. Information sources To comprehensively gather relevant studies, we searched five major scientific databases until April 2024: PubMed, Scopus, Web of Science, IEEE Xplore, and SpringerLink. Team members initially drafted the search strategies and then refined them under the guidance of our supervisor. The final search outcomes were exported to EndNote, and the team carefully removed duplicate entries through a double-check process. 2.3. Search The search strategy was developed using a combination of keywords organized into three categories: Model, Technique, and Disease (Table 1). For the disease category, which focuses on Dysgraphia, we also included relevant MeSH [26] terms to ensure comprehensive coverage. We employed the ‘AND’ operator between different categories and the ‘OR’ operator within the same category to effectively refine and broaden our search results as needed. 10.1371/journal.pone.0328722.t001Table 1List of keywords applied in the search strategy.Model-Related WordsTechnique-Related WordsDisease-Related WordsAI/Artificial Intelligence/ML/Machine Learning/DL/Deep Learning/Image ProcessingCNN/Convolution Neural Networks/SVM/Support Vector Machine/NNs/Neural Networks/AdaBoost/Adaptive Boosting/U-Net/UNet/VGGNet/ResNet/Residual Net/Random Forest/Decision Tree/ PCA/LDA/AutoEncoder/GAN/Generative Adversarial Network/GoogleNet/MobileNet/DenseNet/YOLO/RetinaNet/Naïve Bayes/MLP/Multi-Layer PerceptronDysgraphia/Dysgraphias/Argraphia/Argraphias*Mesh terms are bold List of keywords applied in the search strategy. *Mesh terms are bold 2.4. Selection of sources of evidence To ensure alignment with the defined eligibility criteria and maintain consistency among reviewers, all titles and abstracts of publications were double-screened by two reviewers. In cases of unclear points, both reviewers screened the full text. Any disagreements regarding inclusion were resolved through discussions with the supervisor until a consensus was achieved. The same two reviewers conducted full-text screening to extract the features. All data were managed using a spreadsheet environment. 2.5. Data charting process The data charting process was meticulously conducted using a structured form developed in Google Sheets. This form included columns for key study characteristics, datasets, methods, and results. The selection of items for charting was based on previously established eligibility criteria and reviewed by two reviewers. One of the reviewers participated in the data charting process, while the other independently charted each study. Discrepancies were resolved through thorough discussions. Inconsistencies were addressed during team meetings, and data verification involved confirming details with original study authors when necessary. The charting process was iterative, with the form updated based on initial findings. Additional variables were added after the first round of charting to ensure a comprehensive data collection. Data confirmation was supervised by the supervisor, ensuring accuracy and reliability. 2.6. Data items The review included qualitative and quantitative data, capturing a comprehensive spectrum of study characteristics and outcomes. Quantitative data, such as accuracy rates, were charted alongside qualitative descriptions of study objectives. Interpretation was necessary for categorizing intervention types and understanding contextual factors from study descriptions, particularly for variables like feature extraction methods. The final charting form comprised columns for study characteristics (e.g., author, year), datasets used, methods (e.g., model training, evaluation metrics), and results (e.g., accuracy, precision, strengths). Variables collected included publisher details, dataset features, size, methods of data collection, detailed methodological approaches, and performance metrics. For instance, study characteristics captured information such as publisher impact factor and open access status, while dataset details included features like X/Y coordinates and pen position. This comprehensive approach ensured a thorough synthesis of the available evidence, providing valuable insights into detecting dysgraphia through various methodologies. 2.7. Critical appraisal of individual sources of evidence In our study, we critically appraised the included studies to guarantee the quality and relevance of the evidence. This appraisal was carried out using the Critical Appraisal Skills Programme (CASP) [27] checklists for qualitative studies and the Newcastle-Ottawa Scale [28] for quantitative studies. Each study was evaluated based on its methodological rigor, clarity of reporting, and relevance to the research question, focusing on dysgraphia diagnosis using handwritten images. The CASP checklist facilitated a structured review of qualitative studies, assessing aspects such as the validity of the research, clarity in reporting, and appropriateness of the methodology. Studies were scored on key components such as the research aim, design, recruitment strategy, data collection methods, and findings. For quantitative studies, the Newcastle-Ottawa Scale was employed to evaluate the quality of non-randomized studies. It specifically focused on three main criteria: selection of study groups, comparability of groups, and the ascertainment of either the exposure or outcome of interest. Studies were awarded stars for meeting specific criteria within these categories, with a maximum of nine stars indicating the highest quality. 2.7.1. Proposed appraisal method explanation. In this study, we implemented a detailed scoring system to critically appraise the included studies, ensuring a rigorous evaluation of their quality and relevance. This system is based on three main criteria: validity, results, and relevance, with each criterion contributing to an overall score that reflects the study’s robustness and applicability. a)Validity was assessed according to the variety of models applied in the study. Studies that applied more than four models received the highest score of 3 points, those using 3 or 4 models received 2 points, and studies applying fewer than three models received 1 point. This measure reflects the comprehensiveness of the study’s approach to model evaluation and comparison, which is crucial for robust and reliable findings. Validity was assessed according to the variety of models applied in the study. Studies that applied more than four models received the highest score of 3 points, those using 3 or 4 models received 2 points, and studies applying fewer than three models received 1 point. This measure reflects the comprehensiveness of the study’s approach to model evaluation and comparison, which is crucial for robust and reliable findings. b)Results were evaluated by examining the reported accuracy of the study’s outcomes. This measure acknowledges the significance of precise and reliable results in validating a study’s conclusions. Studies reporting an accuracy greater than 90% were awarded 3 points, those between 80% and 90% received 2 points, and studies with accuracy between 70% and 80% received 1 point. Studies reporting accuracy below 70% or not reporting accuracy at all received 0 points. Results were evaluated by examining the reported accuracy of the study’s outcomes. This measure acknowledges the significance of precise and reliable results in validating a study’s conclusions. Studies reporting an accuracy greater than 90% were awarded 3 points, those between 80% and 90% received 2 points, and studies with accuracy between 70% and 80% received 1 point. Studies reporting accuracy below 70% or not reporting accuracy at all received 0 points. c)Relevance was determined by the extent to which the study focused on dysgraphia, handwriting, and classification problems. Studies mentioning “dysgraphia” and “handwriting” received the highest score of 3 points, indicating a direct relevance to our research focus. Studies mentioning only “dysgraphia” received 2 points, while those mentioning only “handwriting” received 1 point. Additionally, studies that addressed a classification problem received 1 point. Studies that did not mention either term or address classification problems received 0 points. Relevance was determined by the extent to which the study focused on dysgraphia, handwriting, and classification problems. Studies mentioning “dysgraphia” and “handwriting” received the highest score of 3 points, indicating a direct relevance to our research focus. Studies mentioning only “dysgraphia” received 2 points, while those mentioning only “handwriting” received 1 point. Additionally, studies that addressed a classification problem received 1 point. Studies that did not mention either term or address classification problems received 0 points. The total score for each study, ranging from 0 to 9, was calculated by summing the points from all three criteria. This total score allowed us to classify the studies into categories of high quality (7–9 points), moderate quality (4–6 points), and low quality (0–3 points). This structured approach ensured a comprehensive and transparent appraisal of the evidence, allowing us to base our conclusions on the most reliable and relevant studies available. 3. Results 3.1. Selection of sources of evidence The initial search results were collected and consolidated using EndNote [29], with duplicates removed (Fig 2). The screening process and data selection, as outlined in Fig 2, were based on PRISMA guidelines [30]. From an initial extraction of 177 papers, 135 remained after duplicate removal. Abstracts were reviewed to exclude studies not primarily focused on dysgraphia, such as those diagnosing Alzheimer’s and Parkinson’s disease through handwritten images or those that were review papers. Additionally, studies employing diagnostic methods like EEG or brain scans were set aside, keeping only those involving handwritten images. Applying these inclusion and exclusion criteria reduced the number of relevant papers to 37. The review process advanced with a thorough examination of the complete texts, further excluding those that were not primarily related to dysgraphia or did not conform to models based on handwritten images, resulting in 21 retained papers. This study is a scoping review focused on identifying papers that diagnose dysgraphia using computational methods. Overview of study selection using the PRISMA framework. 3.2. Literature context and research gaps The publication dates span nearly a decade, from 2015 to the present (Fig 3). Of the 21 selected papers, 8 (n = 8, 38%) [31–37] are conference papers, and the remaining 13 (n = 13, 62%) [20,38–57] are journal articles. The papers originate from various countries, including India, Qatar, Spain, Switzerland, Slovakia, the Czech Republic, Israel, France, the USA, Malaysia, Morocco, and Austria. Additionally, some papers are collaborative efforts involving researchers from multiple nationalities. In addition, only nine of the included studies are open-access (n = 9, 42%) [20,38–40,43–45,48], and the 12 other papers are private (n = 12, 58%) [31–33,41,42,34–37,46–47,58]. General characteristics of the included studies. A. Distribution of journal articles versus conference proceedings; B. The number of publications by year shows the publication trends from 2015 to 2024; C. Geographic distribution of studies by country; D. Proportion of public versus private access to studies. While our review provides detailed insights into the current AI methodologies for diagnosing dysgraphia, it is essential to contextualize this within the broader academic dialogue. Previous reviews have provided foundational knowledge in this field; however, most adopt a clinical perspective, prioritizing diagnostic criteria and behavioral assessments over AI’s potential to transform dysgraphia diagnosis [49,50]. Moreover, few are dedicated solely to dysgraphia, often embedding it within broader discussions of specific learning disabilities (SLDs) like dyslexia and dyscalculia, thus diluting focus on dysgraphia-specific challenges [51–53]. Even reviews exploring AI applications, such as [54,55], typically emphasize traditional machine learning methods (e.g., SVM, AdaBoost) and overlook state-of-the-art approaches, notably the integration of traditional and modern techniques, such as combining handwriting image analysis with dynamic features like pen pressure or motor patterns, which our review examines in detail. Additionally, many prior reviews are outdated, failing to reflect recent AI advancements. One of the more comprehensive recent works is the review by Kunhoth et al. (2024) [56], which provides a broad overview of various learning disorders and explores both machine learning and non-machine learning approaches for their prediction. While their work offers valuable general insights and effectively groups and compares different models, it lacks the depth of analysis found in our study. In contrast, our review includes more specific technical details, especially about the used datasets, considers a wider array of models, and delves into how they are implemented. It also provides deeper guidance and recommendations, making it more actionable for researchers and practitioners. Together, these additions make our review not only complementary to previous works but also a necessary update for the current state of AI research in this domain. Our work addresses these gaps by providing a dysgraphia-specific, up-to-date synthesis of AI applications, comprehensively covering both traditional and cutting-edge methods. This contribution highlights the multidisciplinary relevance of AI in dysgraphia diagnosis through handwritten image analysis, underscoring its potential impact across educational and healthcare settings. Despite growing interest, a recent slowdown in publications suggests a bottleneck in data availability, emphasizing the need for improved data collection and sharing to sustain research momentum in this critical area. 3.3. Specific characteristics of included papers We extracted information regarding the dataset, algorithm and methodology, and performance results and analyzed each aspect individually. 3.3.1. Dataset characteristics. The analyzed data in the included papers were sourced from various repositories such as databases, registers, and health information systems (Table 2). Information about datasets was extracted in both general and specific ways. Multiple tasks were applied to gather the dataset, including copying sentences, writing digits, writing individual words, and performing the BHK test in communication with humans and robots. Two papers utilized multiple datasets (n = 2, 9%) [20,33], while only three used raw handwritten images (n = 3, 14%) [32,37,57], including words, sentences, and digits. The other studies incorporated additional features, including kinematic and temporal characteristics, and documented supplementary data captured during tasks using specialized equipment such as Wacom Tablets. Open-access datasets are provided in Table 3. 10.1371/journal.pone.0328722.t002Table 2Summary of dataset information.StudyTaskFeaturesAgeLanguageLabelsAccessParticipantsTotal size[20,31,38–40]Letters & Sentences: varying speedSpatial/Temporal/Dynamic15-AugSlovakDysgraphia/NormalOpen120: 57D480[32]LettersSpatial/Dynamic––Dysgraphia/NormalOpen–267930[41]–Spatial/Temporal/Dynamic–SlovakDysgraphia/NormalClose––[58]Digits & Letters & OtherKinematic/Error AnalysisSchool-going childrenSinhala5 Series of labelsOpen65713353 Sentences & 115320 Words[33]LettersKinematic/Quality/Shape Features–FrenchNormalOpen–810000 characters[33]–Kinematic/Quality/Shape FeaturesSchool-going childrenFrenchDysgraphiaClose––[42]Think and writeSpatial/Quality/Error Analysis/OthersGrade 1–7–Dysgraphia/NormalClose21003131[43]BHK TestSpatial/Dynamic/MovementGrade 2–5FrenchDysgraphia/Normal–580: 120D580[44]BHK TestSpatial/Dynamic/DemographicPrimary SchoolFrenchDysgraphia/Normal–448:58D448[34]Copy a paragraphKinematic/Temporal/Spatial/Dynamic/TQWT-derived featuresGrades 3 & 4CzechDysgraphia/Normal–97–[35]Repeating LettersKinematic/Pen Dynamics/Others10–13SlovakDysgraphia/Normal–78:48D–[36]ArchimedeanSpiral repetitive loops/OthersKinematicAverage of 69P and 62NCzechParkinson/Normal–69:33P–[45]Copying TextStatic/Kinematic/Pressure/TiltPrimary SchoolFrenchDysgraphia/Normal–298:56D298[46]3 Series of tasksTemporal/Pressure/Spatial/On-Paper/OthersGrade 3HebrewDysgraphia/Normal–99:49D–[37]Copying TextSpatial/OtherGrade 1–7KannadaRandom Students––940[47]LettersKinematic/Geometric/Non-Linear Dynamic/Tablet Specific Signal/OthersGrade 3HebrewDysgraphia/Normal–54:27D54[48]Letters & WordsHandwritten Images8-JunIndianDysgraphia/NormalOpen150–[57]Writing Digits with Interactive RobotHandwritten Digits12-JunLatinDysgraphia/NormalOpen17414004 Summary of dataset information. 10.1371/journal.pone.0328722.t003Table 3Open access datasets.StudyDatasetLink[20,31,38–40,48]Dysgraphia Detection Through Machine Learning https://github.com/peet292929/Dysgraphia-detection-through-machine-learning [32]Dysgraphia Handwriting SampleContact Association of Dyslexia Malaysia (ADM) [50][58]IAM Handwriting Database https://fki.tic.heia-fr.ch/databases/iam-handwriting-database [57]Handwritten Digits https://ijece.iaescore.com/index.php/IJECE Open access datasets. 3.3.2. Studies’ methodology characteristics. Various models have been applied in these studies (Table 4), including SVM and Random Forest (RF), which were used as benchmarks to compare with the innovative model in a considerable number of studies (SVM = 11, 52% - RF = 9, 42%). One study deviated from the norm by employing a clustering method instead of a traditional classifier, utilizing it as a classifier (n = 1, 4%). The primary evaluation metric is accuracy used by 19 studies (n = 19, 90%), ranging between 79% and 99%. Additionally, three studies used ROC and AUC metrics (n = 3, 14%), and three employed the Matthews correlation coefficient (n = 3, 14%). 10.1371/journal.pone.0328722.t004Table 4Overview of methodological approaches.StudyModelsFeature ExtractionFusion ModelData AugmentationNoise RemovalNormalizationBinarizationDimensionality ReductionPerformance[38]AdaBoost/SVM/RFWKNN_F5–Zooming/Shifting/Shearing–Z-Score––ACC 79.7/SPEC 76.7/SEN 79.7[32] Directed Acyclic Graph Convolutional Layers–Rotation/Resizing–Batch Normalization–Average PoolingACC 99/ PRE 98/ REC 99/ F1 99[41]CNN/SVM/RF/NB/DT/KNNConvolutional Layers–Flipping–Min Max Scaling–Max PoolingACC 91/PRE 91/REC 91/ F1 95[39]KNN/SVM/RF/AdaBoostComprehensive Extraction–––Min Max Scaling–BoostACC 80/ PRE 83/ REC 78/ F1 80/ ROC & AUC[58]CNN: RNN + LSTMConvolutional Layers–Gaussian Blur/ Dilation/ Erosion––TRUE–ACC 73[20]SVM/RF/AdaBoostDenseNet201Horizontal Concatenation: All pairs & TriadsZooming/Shifting/Shearing––––ACC 97/ PREC 97/ REC 97/ F1 97/ AUC & ROC[33]CNN/ResNet/VGG16/Inception V3Convolutional Layers–Mirroring–Resizing–Inception V3ACC 98/ SEN 95/ SPEC 93[42]RF/SVM/LR/One-Class SVMContour Analysis/Average non-zero pixels––Bilateral Filtering–TRUE–ACC 92/ PREC 90/ REC 77/ F1 83[40]CNN/AdaBoost/SVM/RFInterpolation/DFT–––Interpolation––ACC 79/TPR 76/ TNR 80[43]Gaussian Process/SVM/LSTM/DT/RF/NB/AdaBoost/Gradient Boost/Fisher Criterion/SFFS––Butterworth FilterZ-Score––ACC 86/SPEC 91/ SEN 81/[44] K-means (Binary) Divide into Bins/Signal Extraction/Fourier Transform––Butterworth FilterZ-Score–PCASEN 91/ SPEC 90[34]RF/SVMTQWT––SNRVertical Normalized Jerks–mRMRSEN 80/ SPEC 68/MCC 49[35]RF/SVM/AdaBoostComprehensive Extraction–––Z-Score––ACC 76/ SEN 75/ SPEC 76[36] RF GranWald-Letnikov Approximation––––––ACC 89/ SEN 88/ SPEC 90[45] RF Divide into Bins/ Fourier Transform––Gaussian Filter/ Fourier Transform/Moving Average FilterAveraging Fourier Transforms––SEN 96/ SPEC 99/ F1 97[46] SVM Pen Displacement Analysis––Hann Kernel FilteringZ-Score––ACC 89/ SPEC 90/ SEN 90/ ROC & AUC[37]MobileNet/CNNBounding Boxes and Contours––Gaussian BlurContrast & Brightness Normalization––Classifying into Below Ave/ Ave/ Above Ave[31]AdaBoost/SVM/RF/KNNComprehensive Extraction––––––ACC 79[47]RF/LDAComprehensive ExtractionMultidimensional Fusion Model/HPSQ––Cohort Normalization–mRMR/SFFSACC 96/ SPEC 96/ SEN 96/ TSS 1.99[48]NDR-R2CNN/SVM/RF/DTConvolutional Layers–Rotating/ScalingGaussian BlurZ-ScoreTRUE–ACC 98/ PREC 96/ REC 100/ F1 98[57] CNN Convolutional Layers––––––ACC 91/ PREC 93/ REC 91/ F1 91* Models in BOLD are the main model* Short Forms: CNN: Convolutional Neural Network/DT: Decision Tree/RF: Random Forest/LDA: Linear Discriminant Analysis/LR: Logistic Regression/AdaBoost: Adaptive Boosting/KNN: K Nearest Neighbors Overview of methodological approaches. * Models in BOLD are the main model * Short Forms: CNN: Convolutional Neural Network/DT: Decision Tree/RF: Random Forest/LDA: Linear Discriminant Analysis/LR: Logistic Regression/AdaBoost: Adaptive Boosting/KNN: K Nearest Neighbors 3.4. Results of critical appraisal within sources of evidence The critical appraisal of the included studies revealed a range of scores spanning from 3 to 9 (Table 5, Fig 4). Three papers achieved the highest score of 9, demonstrating exceptional quality. Four papers scored 8, and five scored 7, indicating high quality. Six papers scored 6, suggesting moderate quality, while two papers scored 5. Notably, no papers received a score of 4, and one paper scored 3, reflecting lower quality. This distribution shows that most studies are of high to moderate quality, underscoring the robustness of the evidence base. 10.1371/journal.pone.0328722.t005Table 5Results of scoring.StudyFirst AuthorYearValidity ScoreRelevance ScoreResult ScoreTotal ScoreCategory[38]Peter Drotar20202316Moderate Quality[32]Siti Azura Ramlan20231337High Quality[41]Richa Gupta20233339High Quality[39]Jayakanth Kunhoth20232327High Quality[58]Neha N. Doshi20231315Moderate Quality[20]Jayakanth Kunhoth20232338High Quality[33]Sharmila C20232338High Quality[42]Basant Agarwal20233339High Quality[40]Juraj Skunda20222316Moderate Quality[43]Louise Deschamps20213328High Quality[44]Thibault Asselborn20201337High Quality[34]Vojtech Zvoncak20191326Moderate Quality[35]Zuzana Dankovicova20192316Moderate Quality[36]Jan Mucha20182215Moderate Quality[45]Thibault Asselborn20181337High Quality[46]Sara Rosenblum20171326Moderate Quality[37]Tushar B. T.20241203Low Quality[31]Jayakanth Kunhoth20222316Moderate Quality[47]Jiri Mekyska20171337High Quality[48]Fatima Ghouse20213339High Quality[57]Soukaina Gouraguine20231337High Quality Results of scoring. Quality assessment scores of included studies. 3.5. Comparison of mostly used predictive models This section will discuss the most widely applied predictive models of dysgraphia detection featured in the included papers: SVM, KNN, RF, K-means, AdaBoost, and CNN. We aim to provide a comprehensive analysis by considering both their strengths and weaknesses. This balanced examination will help us understand each model’s effectiveness and limitations, thereby guiding the selection of the most appropriate techniques for our research objectives. 3.5.1. SVM. SVM [59] is a classification model designed to identify the optimal hyperplane that separates data points in the feature space while maximizing the distance between the closest points on either side of the hyperplane (Fig 5). This margin maximization produces a robust model, as demonstrated in linearly separable data examples. Examples of linearly separable data distributions [60]. In ideal scenarios, data are linearly separable, allowing for straightforward training of a linear SVM. However, real-world data often requires a more flexible approach. By introducing slack variables, SVMs can maximize the soft margin, permitting some classification errors and thereby improving the handling of nearly linearly separable data and reducing overfitting. For linearly inseparable data, SVMs utilize kernel functions to map data to higher-dimensional spaces, making them linearly separable. Eleven of the included papers (n = 11, 52%) [20,31,34,38–43,46,48] utilized SVM for processing handwritten images, integrating additional features such as kinematic, temporal, and spatial data. The most effective kernels employed were the one-class SVM [42] and the Radial Basis Function (RBF) kernel [43]. Additionally, combining SVM with fusion models has been shown to be particularly effective in improving the performance of these systems [20]. In the context of dysgraphia diagnosis, SVMs excel in handling small, high-dimensional handwriting datasets, such as those incorporating kinematic, spatial, and temporal features from Wacom Tablets [20,43]. Their robustness to overfitting makes them particularly suitable for language-specific datasets with limited samples, as demonstrated by the use of one-class SVM and RBF kernels [42,43]. However, their high computational cost may pose challenges for real-time applications in educational or clinical settings, and performance relies on careful kernel selection to accommodate the variability in dysgraphic handwriting patterns. A summary of the advantages and disadvantages of the SVM algorithm for dysgraphia diagnosis is provided in Table 6. 10.1371/journal.pone.0328722.t006Table 6Advantages and disadvantages of the SVM algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesEffective for small, high-dimensional handwriting datasetsHigh computational cost and time consumption on large-scale datasets limit real-time useRobust to overfitting in limited-sample, language-specific tasksKernel choice sensitivity affects performance on diverse handwriting Advantages and disadvantages of the SVM algorithm in dysgraphia diagnosis. 3.5.2. KNN. The K-Nearest Neighbors (KNN) algorithm classifies data points by their closeness to other data points within the feature space [61]. For a given input, it identifies the K closest neighbors and assigns the most common label among them to the input. This method assumes data points with similar characteristics are located near each other, making it practical for various classification tasks, including image processing (Fig 6). KNN is non-parametric and straightforward to implement but can be computationally intensive with large datasets and sensitive to the choice of K and distance metrics. KNN algorithm overview [62,63]. In four of the included papers (n = 4, 19%) [38,39,31,41], KNN was utilized; in three of these, it served as a classifier, while in one, a variant called Weighted KNN_F5 [64] was used for feature extraction. Weighted KNN_F5 assigns different weights to neighbors based on their distance, giving closer neighbors more influence on the classification outcome. While the results of KNN as a classifier were not particularly strong, Weighted KNN F5 for feature extraction showed the potential to enhance the model’s performance. For dysgraphia diagnosis, KNN’s simplicity makes it a valuable baseline in early-stage studies, effectively classifying spatial and temporal handwriting features [38,39]. Variants like Weighted KNN F5 enhance feature extraction, improving performance with small datasets [64]. However, its sensitivity to noise, such as inconsistent stroke patterns in dysgraphic handwriting, limits its accuracy in complex cases [31]. Additionally, KNN’s computational intensity with large or high-dimensional datasets poses challenges for scalable dysgraphia screening in educational settings. A summary of the advantages and disadvantages of the KNN algorithm for dysgraphia diagnosis is provided in Table 7. 10.1371/journal.pone.0328722.t007Table 7Advantages and disadvantages of the KNN algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesSimple baseline for early dysgraphia detection due to its simplicity and non-parametric natureSensitive to handwriting noise, reducing accuracy in complex dysgraphia tasksWeighted variants improve performance on small datasets, benefiting from low training timeHigh cost on large/high-dimensional data due to memory usage, limiting real-world scalability Advantages and disadvantages of the KNN algorithm in dysgraphia diagnosis. 3.5.3. Random forest. The Random Forest algorithm [65] is a versatile and robust ensemble learning technique for classification and regression tasks. It works by constructing multiple decision trees [66] during training and outputting the mode of the classes (for classification) or mean prediction (for regression) of the individual trees (Fig 7). This approach leverages the wisdom of crowds, resulting in more accurate and robust predictions compared to a single decision tree. Random forest process [67]. Nine of the included papers (n = 9, 42%) [34–36,38,39,32,41,42,47] applied the Random Forest (RF) algorithm for dysgraphia detection based on images. While RF showed promising results when combined with other models, its performance was not as strong. In dysgraphia detection based on images, Random Forest is particularly effective because it can manage high-dimensional data and capture complex patterns [36,45]. In dysgraphia diagnosis, Random Forest’s ability to handle mixed feature types, such as spatial, temporal, and pressure-based features, makes it highly effective for multimodal handwriting analysis, achieving robust performance in complex datasets, particularly valuable for diverse handwriting patterns in dysgraphia studies [36,45]. However, its computational intensity may hinder real-time screening in clinical or educational settings, and performance can vary with imbalanced datasets, necessitating careful feature selection [36]. A summary of the advantages and disadvantages of the Random Forest algorithm for dysgraphia diagnosis is provided in Table 8. 10.1371/journal.pone.0328722.t008Table 8Advantages and disadvantages of the random forest algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesHandles mixed feature types and high-dimensional data wellComputationally intensive for large datasets; limits real-time screeningBoosts accuracy in multimodal handwriting analysisPerformance varies with imbalanced data; needs careful feature selection and may overfit without proper tuning Advantages and disadvantages of the random forest algorithm in dysgraphia diagnosis. 3.5.4. K-means. K-means is an unsupervised machine algorithm that groups data into clusters to identify distinct classes [68]. The algorithm involves four main steps: First, K initial clustering centers are selected. Second, each sample in the dataset is allocated to the nearest cluster center based on distance calculations. Third, the cluster centers are recalculated for each group. Finally, the second and third steps are repeated until a specified stopping criterion, such as reaching the maximum number of iterations or achieving minimal changes in cluster centers, is met (Fig 8). The example of the K-means algorithm [69]. K-means can also be adapted for classification tasks by defining the number of clusters, K, to correspond to the number of classes [44]. After the clustering process, new data points can be classified by assigning them to the nearest cluster center. This approach leverages the simplicity and efficiency of K-means to create a basic classification model that groups data points based on their feature similarity to predefined cluster centers. In dysgraphia diagnosis, K-means is effective for clustering dysgraphic vs. non-dysgraphic handwriting patterns in exploratory studies, leveraging its simplicity for small datasets [44]. Its computational efficiency supports initial detection tasks. However, its performance is limited by small, imbalanced dysgraphia datasets, which may result in poor clustering of complex handwriting patterns [44]. Additionally, the need to predefine the number of clusters poses challenges for heterogeneous dysgraphia data with varied writing styles. A summary of the advantages and disadvantages of the K-means algorithm for dysgraphia diagnosis is provided in Table 9. 10.1371/journal.pone.0328722.t009Table 9Advantages and disadvantages of the K-means algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesEffectively clusters dysgraphic vs. non-dysgraphic patternsStruggles with imbalanced data and outliers; poor clustering of complex patternsSimple, efficient, and easy to interpret for small handwriting datasets due to low complexityRequires predefined cluster numbers; limited with heterogeneous or highly discrete classes Advantages and disadvantages of the K-means algorithm in dysgraphia diagnosis. 3.5.5. AdaBoost. AdaBoost, short for Adaptive Boosting, is a powerful ensemble learning algorithm for classification tasks, introduced by Yoav Freund and Robert Schapire in 1996 [70]. AdaBoost’s core principle is to improve weak classifiers’ accuracy by integrating them into a robust, strong classifier (Fig 9). It starts by assigning equal weights to all training samples and iteratively trains a sequence of weak classifiers, typically simple models like decision stumps. After each iteration, it evaluates the classifier’s error rate and adjusts the sample weights, increasing the weights of misclassified samples to ensure subsequent classifiers focus on harder-to-classify cases. The final model is constructed by aggregating all the weak classifiers, where each classifier’s influence is proportional to its accuracy. AdaBoost is known for its ability to significantly enhance model performance, its adaptability to various types of weak classifiers, and its robustness against overfitting, although it can be sensitive to noisy data and outliers. This adaptive focusing mechanism and iterative process make AdaBoost a powerful tool in machine learning for creating highly accurate classification models. Notably, seven of the included papers utilized AdaBoost (n = 7, 33%) [20,31,38–40,43,35], with three of them (n = 3, 14%) employing it as the primary model in their studies [38,39,47]. For dysgraphia diagnosis, AdaBoost enhances the accuracy of weak classifiers, making it effective for small handwriting datasets with diverse features like spatial and kinematic attributes [20,38,39]. Its robustness to overfitting supports its use in combining multiple handwriting features. However, its sensitivity to noisy data, such as irregular strokes in dysgraphic samples, may reduce performance in complex cases [20]. Additionally, the need for careful tuning of weak classifiers increases complexity for practical dysgraphia applications. A summary of the advantages and disadvantages of the AdaBoost algorithm for dysgraphia diagnosis is provided in Table 10. 10.1371/journal.pone.0328722.t010Table 10Advantages and disadvantages of the AdaBoost algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesBoosts weak classifiers, improving accuracy in small, diverse handwriting datasetsSensitive to noisy handwriting; lowers performance in complex casesAdaptRobust to overfitting; adapts well to combined handwriting featuresRequires careful tuning; adds computational complexity Advantages and disadvantages of the AdaBoost algorithm in dysgraphia diagnosis. The example of the AdaBoost process [71]. 3.5.6. CNN. Convolutional Neural Networks [72] are a type of deep learning algorithms [73] particularly well-suited for image processing tasks. CNNs are designed to automatically and adaptively learn hierarchical spatial features from input images. The architecture of a CNN typically consists of several layers, including convolutional layers, pooling layers, and fully connected layers (Fig 10). The convolutional layers perform convolution operations to the input image, using a set of learnable filters to produce feature maps that capture different input characteristics, such as edges, textures, and patterns. These feature maps are then passed through activation functions to introduce non-linearity, enhancing the model’s ability to learn complex representations. Pooling layers follow, performing down-sampling operations that shrink the spatial dimensions of the feature maps, thereby reducing computational requirements and providing some degree of translation invariance. Lastly, fully connected layers integrate the extracted features to perform the final classification or regression task. This hierarchical structure allows CNNs to learn more abstract and high-level features progressively, making them very effective for tasks such as image classification, object detection, and segmentation. Sample of a typical CNN [74]. Seven of the included papers (n = 7, 33%) utilized CNNs, with two (n = 2, 9%) specifically employing well-known CNN-based models: one of the papers [37] using MobileNet [75] and the other paper [33] collectively incorporating Inception [76], VGGNet [77], and ResNet [78]. These studies achieved a high level of accuracy, with an average accuracy of 88% across all papers that used CNNs. Although CNNs are limited in capturing long-distance dependencies [52], they performed exceptionally well in most cases presented in the included papers. In dysgraphia diagnosis, CNNs excel in learning complex patterns from handwriting images, achieving high accuracy with models like MobileNet and ResNet [33,37]. Their ability to analyze spatial and kinematic features makes them ideal for robust image-based diagnosis. However, their reliance on large, diverse datasets is a challenge given the small sample sizes common in dysgraphia research [33]. Additionally, high computational demands may limit their use in resource-constrained educational or clinical settings. A summary of the advantages and disadvantages of CNNs for dysgraphia diagnosis is provided in Table 11. 10.1371/journal.pone.0328722.t011Table 11Advantages and disadvantages of the CNN algorithm in dysgraphia diagnosis.AdvantagesDisadvantagesComputationally efficient with rapid training and high accuracyRequires large, diverse handwriting datasets, often limited in dysgraphia researchExcels at learning complex handwriting patterns and spatial featuresLess effective at capturing long-distance features, possibly missing global patternsHigh computational demands may limit use in resource-constrained settings Advantages and disadvantages of the CNN algorithm in dysgraphia diagnosis. Overall, the choice of an AI model for dysgraphia detection depends on dataset characteristics and application constraints. The trade-offs discussed highlight the importance of balancing model complexity, computational resources, and dataset limitations in achieving effective dysgraphia diagnosis. 3.6. Feature analysis Feature selection is pivotal to the effectiveness of AI-based models for dysgraphia diagnosis. The quality, type, and modality of features extracted from handwriting samples significantly influence model accuracy and diagnostic reliability. This section analyzes features utilized in the reviewed studies, categorizing them by type, detailing their application in online and offline handwriting modalities, and evaluating their impact on model performance with specific examples. 3.6.1. Categorization of features. AI models for dysgraphia leverage diverse features from handwriting samples, classified into four categories. Each category’s role and application differ between online (real-time data, e.g., tablet-based writing) and offline (static data, e.g., scanned images) handwriting, reflecting distinct diagnostic insights. 3.6.1.1. Temporal features: Temporal features capture timing aspects of handwriting, such as stroke duration, inter-stroke pauses, and total writing time. In online handwriting, these features are directly measured via stylus or tablet devices, providing precise data on rhythm and pacing irregularities—hallmarks of dysgraphia. For example, temporal features can detect hesitations in letter formation, which are critical for identifying dysgraphic children who struggle with writing fluency. In offline handwriting, temporal features are inferred indirectly from stroke patterns or writing density, but their accuracy is limited due to the absence of real-time data. These features are particularly valuable in online settings for early detection, as they reveal subtle motor control issues not visible in static samples [38]. 3.6.1.2. Spatial features: Spatial features encompass geometric characteristics, including letter size, shape, alignment, spacing between characters/words, and line adherence. In offline handwriting, spatial features are primary, extracted from scanned images to identify visual irregularities like inconsistent letter sizes or crowded spacing, common in dysgraphia. These features are robust for static analysis, as they rely on visible patterns rather than dynamic input. In online handwriting, spatial features are less dominant but still relevant, capturing real-time stroke geometry. Spatial features play a central role in both modalities for diagnosing dysgraphia by highlighting structural handwriting deficits, though offline applications are more prevalent due to their reliance on image-based data [79]. 3.6.1.3. Pressure-based features: Pressure-based features measure the force exerted on the writing surface, including mean pressure, pressure variability, and peak force points. In online handwriting, these are captured directly by pressure-sensitive devices, revealing fine motor skill deficits (e.g., erratic pressure patterns indicating grip issues in dysgraphia). Pressure features are critical for assessing motor control precision, a key diagnostic marker. In offline handwriting, pressure features are rarely used, as scanned images lack force data, though some studies infer pressure from stroke thickness. These features are thus more significant in online modalities, enhancing models’ ability to detect motor-related dysgraphia symptoms [31]. 3.6.1.4. Kinematic features: Kinematic features quantify handwriting motion, including writing speed, acceleration, jerk (rate of acceleration change), and stroke smoothness. In online handwriting, kinematic features are richly captured via real-time tracking, enabling precise analysis of movement fluidity (e.g., jerky strokes in dysgraphic writers). These features are essential for detecting motor coordination issues. In offline handwriting, kinematic features are inferred from stroke trajectories or curvature, but their granularity is reduced due to static data limitations. Kinematic features are vital in both modalities for assessing handwriting dynamics, with online applications offering superior detail for early diagnosis [39,31,79]. 3.6.2. Trends in feature selection. The selection of features in AI models often depends on the study’s specific objectives and the nature of the data available. Trends observed across the reviewed studies include: a)Online Handwriting: Studies prioritize temporal and pressure-based features to capture real-time dynamics, such as irregular pauses or inconsistent pressure. For example, one study used temporal features to detect prolonged stroke durations, while another leveraged pressure variability to identify motor deficits [31]. Online Handwriting: Studies prioritize temporal and pressure-based features to capture real-time dynamics, such as irregular pauses or inconsistent pressure. For example, one study used temporal features to detect prolonged stroke durations, while another leveraged pressure variability to identify motor deficits [31]. b)Offline Handwriting: Spatial features dominate, focusing on letter shape, size, and spacing to detect visual irregularities. Kinematic features are less common but are used when stroke patterns allow motion inference [79]. Offline Handwriting: Spatial features dominate, focusing on letter shape, size, and spacing to detect visual irregularities. Kinematic features are less common but are used when stroke patterns allow motion inference [79]. c)Hybrid Approaches: Some studies integrate modalities, combining spatial features for offline analysis with temporal and kinematic features for online data, enhancing diagnostic accuracy. Hybrid models are increasingly common in studies aiming for comprehensive dysgraphia detection [21,38]. Hybrid Approaches: Some studies integrate modalities, combining spatial features for offline analysis with temporal and kinematic features for online data, enhancing diagnostic accuracy. Hybrid models are increasingly common in studies aiming for comprehensive dysgraphia detection [21,38]. Additionally, the effectiveness of hybrid models depends on the consistency and precision of data acquisition tools. Variations in tablet sampling rates, resolution, or pressure sensitivity can introduce inconsistencies in feature definitions, impacting the reliability and generalizability of AI models across studies. 3.6.3. Relationship between features and model performance. Feature choice significantly impacts model performance, as demonstrated by reviewed studies: a)Spatial and pressure-based features enhance accuracy in offline handwriting models. For instance, one study [79] achieved 91% accuracy using spatial features (letter size, spacing) in a CNN model, effectively capturing visual dysgraphia markers. Spatial and pressure-based features enhance accuracy in offline handwriting models. For instance, one study [79] achieved 91% accuracy using spatial features (letter size, spacing) in a CNN model, effectively capturing visual dysgraphia markers. b)Temporal and kinematic features improve sensitivity in online handwriting models. One study [31] reported a 15% sensitivity increase in an SVM model by incorporating kinematic features (writing speed), aiding early motor difficulty detection. Temporal and kinematic features improve sensitivity in online handwriting models. One study [31] reported a 15% sensitivity increase in an SVM model by incorporating kinematic features (writing speed), aiding early motor difficulty detection. c)Combining features across modalities boosts diagnostic efficacy. However, inconsistent feature definitions (e.g., varying pressure metrics across devices) hinder cross-study comparisons and standardization [21,38]. Combining features across modalities boosts diagnostic efficacy. However, inconsistent feature definitions (e.g., varying pressure metrics across devices) hinder cross-study comparisons and standardization [21,38]. Furthermore, clinical interpretability of features remains a crucial aspect for real-world adoption. Features that align closely with known dysgraphia symptoms, such as prolonged stroke latency or erratic pressure spikes, can enhance clinician trust and provide actionable insights during assessment. 3.7. Strengths and challenges of studied papers In this section, we examine the strengths and challenges of each reviewed paper, considering both the points explicitly mentioned by the authors and those we identified through our analysis of the papers (Table 12). This comprehensive evaluation clarifies each study’s relative performance and effectiveness, allowing us to determine which approaches performed better on specific datasets. By highlighting both the advantages and limitations, we aim to offer a balanced view that facilitates more informed comparisons and insights into the efficacy of AI-based models in Dysgraphia detection through image processing tasks. 10.1371/journal.pone.0328722.t012Table 12Strengths and challenges of included papers.StudyStrengthsChallenges[38]Comprehensive DatasetSmall Sample Size/Orthography Specific Language: Slovak[32]High Accuracy/Robustness/Great Sample SizeVariability in handwriting patterns among dysgraphia children[41]High Accuracy/Robust Data Augmentation TechniquesPotential variability or bias due to subjective classification criteria[39]Comprehensive Dataset and Extracted FeaturesSmall Sample Size/Orthography Specific Language: Slovak/Combining On-Surface and In-Surface Features[58]High Accuracy/Advanced Techniques of Machine LearningSmall Sample Size[20]Comprehensive Dataset and Extracted FeaturesSmall Sample Size/Orthography Specific Language: Slovak/Combining On-Surface and In-Surface Features[33]High Accuracy/ Use of Advanced Convolutional Models, such as VGG16, ResNet, and InceptionSmall and Variable Dataset[42]Comprehensive Extracted Features/ High AccuracyDifficulty in Data Collection[40]Comprehensive DatasetSmall Sample Size/Orthography Specific Language: Slovak/Combining On-Surface and In-Surface Features[43]Comprehensive Extracted FeaturesSmall Sample Size/Tablet & Software Dependency/BHK Test Subjectivity[44]Low Computational Resource Required/Detecting the Severeness of DysgraphiaSmall & Heterogenous Dataset[34]Improve HD Identification Accuracy by Approximately 20% using TQWT Features, Effective Combination of Conventional and TQWT Features.Small dataset/Subjectivity of Ratings in Dataset/Single Recording Session Limiting Intra-Writing Variability Analysis[35]Comprehensive Extracted Features/Visualization using PCASmall Sample Size/Need for more Advanced Methods for Handwriting Processing and Evaluation.[36]Utilization of Fractional Derivatives in Kinematic Analysis, Achieving high Classification Accuracy with Minimal Features.Small Number of Participants[45]Cost-Effectiveness/High AccuracySubjectivity in Assessment/Small Sample Size[46]High Accuracy/Applicability to Different LanguagesTime-Consuming and Subjective Nature of Current Handwriting Evaluation Methods[37]Utilization of MobileNet for more Efficient Analysis/Specific Focus on Kannada HandwritingOffered Solutions are Time-Consuming/Augmenting the Dataset with Specific Words Children Find Challenging Could Provide a More Targeted Training Dataset[31]Comprehensive Dataset/Multiple Combinations of Features ExaminedTime Consuming/Small Sample Size/ Orthography Specific Language: Slovak[47]Variability in Tasks Done by IndividualsHigh Accuracy/Use of Simple Predictive Model (RF)[48]High Computational Resource NeededHigh Accuracy/Comprehensive Extracted Features/Handled Imbalanced Data[57]Complexity of the Feature Extraction/Time-Costing and Subjectivity of the Data Collection TaskGreat Sample Size/Innovative use of Interactive Robot/ Strengths and challenges of included papers. The primary challenges identified for state-of-the-art models include the high computational resources required, despite achieving accuracies exceeding 90% in all included papers (n = 5, 23%) utilizing modern techniques [20,32,45,47,48]. Another significant issue is the potential limitation in generalizability due to the small sample sizes. Moreover, language-specific models trained to predict particular languages pose a challenge, as their applicability may be limited to those languages only. Subjectivity in data gathering and assessment processes further complicates the evaluation and comparability of results across studies. Conversely, a notable strength of most included papers is the comprehensiveness of the extracted features, which encompass kinematic, spatial, temporal, and other relevant dimensions. This thorough feature extraction enhances the understanding and interpretation of the models’ performance, contributing to more insightful and meaningful conclusions. 4. Discussion Overall, we identified 21 studies on children under 18, especially school-going children, focusing on AI-based models for dysgraphia detection through handwritten image analysis, underscoring the need for early diagnosis and intervention. The review questions focused on identifying the most effective AI-based models, understanding the challenges associated with their application in dysgraphia detection, and suggesting future trends in dysgraphia detection models with higher performance. The evidence gathered highlights that AI models, particularly CNNs and SVMs, are the most commonly used and highly effective for this purpose, with accuracy rates often surpassing 90%. This aligns with the objectives of improving early diagnosis and intervention strategies for dysgraphia. These models effectively utilized both kinematic and spatial features extracted from handwriting samples, demonstrating high accuracy and robustness in classification tasks. Additionally, integrating fusion models and comprehensive feature extraction techniques further enhanced model performance. A more nuanced interpretation of these results shows that while AI models such as CNNs and SVMs offer substantial accuracy, their effectiveness should be contextualized within current clinical practices. Clinicians must consider how these AI tools could mesh with existing diagnostic frameworks to enhance or streamline dysgraphia identification without supplanting necessary human judgment. This consideration is crucial for ensuring that AI complements rather than replaces traditional assessment methods, maintaining a balance between technological advancement and clinical intuition. Critically comparing different AI approaches, CNNs often excel in image-based analysis due to their ability to learn complex patterns from data, making them ideal for analyzing handwriting intricacies. However, SVMs might be preferred in scenarios with fewer data samples, or more straightforward decision boundaries are sufficient, highlighting a trade-off between model complexity and interpretability. Across the studies, methodological limitations such as sample size, diversity, and validation methods have varied, impacting the generalizability of results. These variations suggest a need for standardized research protocols to better compare outcomes across studies. Moreover, potential biases in AI research, such as those introduced by the selection of datasets or the choice of performance metrics, can affect the perceived efficacy of these models. A critical examination of these biases is essential for advancing the field in a meaningful way. Finally, the gap between laboratory findings and real-world applications remains a significant challenge. The controlled environments in which many AI models are developed and tested do not always accurately reflect the complex realities of educational and clinical settings. Bridging this gap requires ongoing collaboration between developers, researchers, and practitioners to ensure these tools are adaptable and practical in everyday use. These findings are particularly relevant to educational institutions, healthcare providers, and policymakers who manage learning disabilities. AI models can facilitate quicker and more accurate diagnoses for educators and healthcare providers, leading to timely and appropriate interventions. For policymakers, the evidence supports the potential for integrating AI-based diagnostic tools into educational and healthcare systems to help individuals with dysgraphia better. 4.1. Limitations 4.1.1. Review process limitations. Several limitations were identified during the scoping review process. The primary limitation was the availability of studies, with only 9 of the 21 included studies being open access. This restriction may have limited the comprehensiveness of the review, but fortunately, we accessed the other 12 studies by emailing the authors for access or purchasing the articles. Additionally, the sample sizes in many studies were small and often specific to certain languages, which may limit the generalizability of the findings. 4.1.2. Extent of information uncovered. The heterogeneity in evaluation metrics, such as accuracy, ROC, AUC, and MCC, across studies complicates direct comparisons of AI model performance in dysgraphia diagnosis. This variability underscores the urgent need for standardized metrics to facilitate consistent and comparable evaluations. Each metric serves distinct purposes in assessing model performance: Accuracy measures overall correctness but can be misleading in imbalanced datasets common in dysgraphia studies, where non-dysgraphic cases often outnumber dysgraphic ones. ROC curves illustrate a model’s diagnostic ability across varying thresholds, AUC quantifies separability between classes, and MCC provides a balanced measure robust to class imbalance, making it particularly valuable for dysgraphia datasets with skewed distributions. Trends in metric usage reveal a preference for accuracy and AUC in studies with large datasets, aiming to balance sensitivity and specificity, while MCC is increasingly adopted in imbalanced scenarios to ensure reliable detection of dysgraphic individuals. In clinical and educational contexts, these metrics have critical implications: high recall (sensitivity) is essential for school-based screening to minimize false negatives, which could delay interventions like occupational therapy for dysgraphic children, whereas high precision is prioritized in clinical settings to reduce false positives, avoiding unnecessary referrals and parental stress. For instance, a CNN-based screening tool deployed in educational platforms (e.g., via web-based systems) might prioritize recall to flag at-risk students, while a diagnostic tool for clinicians would emphasize precision to confirm cases. These choices reflect meaningful trade-offs: false negatives risk delaying support for children in need, while false positives could lead to stress, mislabeling, and misallocated resources. Therefore, selecting metrics aligned with real-world diagnostic objectives is not merely technical—it directly impacts educational outcomes and patient well-being. The implications of metric heterogeneity extend beyond academic challenges; they hinder the synthesis of findings and slow the development of standardized diagnostic models. To address this, future research should adopt a core metric set tailored to specific use cases, supported by public benchmark datasets to ensure fair and consistent comparisons. Best practices such as k-fold cross-validation, threshold tuning, and reporting multiple complementary metrics can further enhance metric reliability. Moreover, the development of interdisciplinary guidelines involving clinicians, educators, and AI researchers can help align evaluation criteria with practical diagnostic needs, ultimately fostering trust and real-world adoption of AI tools in dysgraphia diagnosis. Another challenge lies in the incomplete or unclear information reported in some papers. For instance, certain studies omitted dataset details, such as size, requiring us to infer this from other papers using the same dataset. Additionally, reliance on self-reported data and the subjective nature of some assessments may introduce bias. Furthermore, the search was limited to major databases like PubMed, Scopus, Web of Science, IEEE Xplore, and SpringerLink, potentially missing relevant studies from other sources or unpublished works. Future studies could address this gap through broader, systematic searches to ensure a more comprehensive literature review. 4.1.3. Methodological and research biases. A more nuanced interpretation of the results in light of current clinical practices reveals that while AI models are highly effective under controlled study conditions, their application in typical clinical settings may not always align with traditional diagnostic methods. This misalignment necessitates a careful integration strategy to ensure that AI tools augment rather than replace the clinician’s expertise. A critical comparison of different AI approaches, such as deep learning models like CNNs versus more traditional machine learning models like SVMs, highlights that while CNNs provide higher accuracy in image-based analyses due to their complex pattern recognition capabilities, SVMs offer advantages in terms of computational efficiency and transparency in decision-making. Examining methodological limitations across studies highlights issues such as the small and non-diverse sample sizes, which limit the extrapolation of findings to a broader population. Additionally, many studies do not undergo external validation, raising concerns about the robustness and reproducibility of the findings. Discussion of potential biases in current research points to a prevalence of publication bias, where studies with positive findings are more likely to be published, and selection bias, where the data used may not adequately represent the target population. An analysis of gaps between laboratory findings and real-world application indicates that AI models often perform exceptionally well in controlled environments but may fail to deliver similar results in real-world settings. This discrepancy can be attributed to the complex and variable nature of real-world environments compared to the controlled conditions of most studies. 4.2. Conclusions This scoping review thoroughly examines the current landscape of AI-based models for dysgraphia detection through handwritten image analysis. The results indicate that AI models, particularly advanced machine learning models like CNNs and SVMs, can significantly enhance the diagnostic process for dysgraphia, offering more objective, faster, and more accurate assessments than traditional methods. This is crucial for mitigating the long-term impacts of dysgraphia on educational and social outcomes. This aligns well with the review objectives of identifying effective AI methodologies and understanding their application in real-world settings. Moreover, integrating machine learning techniques with traditional diagnostic methods has enhanced diagnostic accuracy and reliability synergistically. By incorporating machine learning outputs into conventional assessment frameworks, clinicians can better understand dysgraphia characteristics. The analysis of training datasets used in developing AI models revealed a significant variance in dataset quality and size, directly affecting model performance. Our review has identified a pressing need for standardized, large-scale datasets that more accurately reflect the diverse nature of handwriting styles across different demographics. Furthermore, ongoing technological advancements, such as the development of more sophisticated neural network architectures and enhanced computational power, promise to revolutionize the field of dysgraphia diagnosis further. These advancements will likely enable the creation of even more powerful and precise diagnostic tools in the near future. 4.3. Future recommendations Before delving into future recommendations, it is essential to outline the primary roadmap for constructing a model for dysgraphia detection (Fig 11). The review also highlights the need for further research to address the limitations identified. Future studies should focus on issues such as more extensive and more diverse datasets, the development of language-independent models, and the standardization of evaluation metrics. The main steps to follow are listed below: Roadmap of dysgraphia detective model construction. a)One of the most notable strategies is the combination of individual fusion models with traditional machine learning models. This approach has shown remarkable potential, as evidenced by two of the included studies, which achieved impressive accuracy rates of 97% [20] and 96% [47], respectively. This synergy between fusion and traditional models leverages the strengths of both methods, enhancing overall performance and robustness. Integration is typically achieved by extracting handcrafted or statistical features through traditional preprocessing pipelines, such as kinematic analysis or pressure curve interpretation, and feeding them alongside fused deep features into conventional classifiers like SVMs or Random Forests. While this hybrid design can capitalize on domain-specific insights and model generalizability, it may face limitations in scalability, increased computational complexity, and the challenge of optimally weighting heterogeneous feature sources [5,21,80,81]. In addition to developing new models, there is a critical need to enhance the robustness of machine learning integration with traditional diagnostic tools. This can be achieved by refining the algorithms that process the input from these tools, improving the synergy between AI models and existing diagnostic methods. One of the most notable strategies is the combination of individual fusion models with traditional machine learning models. This approach has shown remarkable potential, as evidenced by two of the included studies, which achieved impressive accuracy rates of 97% [20] and 96% [47], respectively. This synergy between fusion and traditional models leverages the strengths of both methods, enhancing overall performance and robustness. Integration is typically achieved by extracting handcrafted or statistical features through traditional preprocessing pipelines, such as kinematic analysis or pressure curve interpretation, and feeding them alongside fused deep features into conventional classifiers like SVMs or Random Forests. While this hybrid design can capitalize on domain-specific insights and model generalizability, it may face limitations in scalability, increased computational complexity, and the challenge of optimally weighting heterogeneous feature sources [5,21,80,81]. In addition to developing new models, there is a critical need to enhance the robustness of machine learning integration with traditional diagnostic tools. This can be achieved by refining the algorithms that process the input from these tools, improving the synergy between AI models and existing diagnostic methods. b)Another critical area for future work is addressing the limitation of inadequate datasets, which significantly impacts the training and evaluation of AI-based dysgraphia detection models. Many included studies relied on small or improperly structured datasets, often limited to specific languages, hindering model generalizability. To overcome this, advanced data augmentation methods can be used to enrich the datasets and enhance model performance substantially. In this review, seven papers utilized data augmentation techniques, achieving an average accuracy of 91%. The most effective augmentation methods identified include shearing, shifting, mirroring, and flipping. These techniques are particularly well-suited for handwriting data in the context of dysgraphia, as they simulate natural variations in writing style, stroke pressure, and spatial alignment—all of which are critical indicators in dysgraphia detection [20,21]. Another critical area for future work is addressing the limitation of inadequate datasets, which significantly impacts the training and evaluation of AI-based dysgraphia detection models. Many included studies relied on small or improperly structured datasets, often limited to specific languages, hindering model generalizability. To overcome this, advanced data augmentation methods can be used to enrich the datasets and enhance model performance substantially. In this review, seven papers utilized data augmentation techniques, achieving an average accuracy of 91%. The most effective augmentation methods identified include shearing, shifting, mirroring, and flipping. These techniques are particularly well-suited for handwriting data in the context of dysgraphia, as they simulate natural variations in writing style, stroke pressure, and spatial alignment—all of which are critical indicators in dysgraphia detection [20,21]. Equally important is the need to enhance dataset diversity across demographics, languages, and cultural contexts to ensure models are robust and equitable. Future studies should prioritize multi-center collaborations involving educational and healthcare institutions worldwide to collect handwriting samples from diverse populations, including varied age groups, genders, socioeconomic backgrounds, and educational levels. Including non-Latin writing systems, such as Arabic, Devanagari, or Chinese, is crucial, as handwriting characteristics vary significantly across scripts [9,44]. For example, partnerships with organizations like the Association of Dyslexia Malaysia could provide access to region-specific samples, while open-access repositories, such as the IAM Handwriting Database, can promote data sharing and inclusivity [45]. Cultural influences, such as writing conventions shaped by regional educational practices, should also be considered to capture context-specific handwriting variations [46,47]. These efforts will reduce biases inherent in language-specific or demographically limited datasets, improving model applicability in diverse clinical and educational settings globally [21,31]. By combining diverse datasets with augmentation techniques, future models can achieve higher generalizability, supporting equitable dysgraphia diagnosis and aligning with the need for standardized, inclusive diagnostic tools. c)Additionally, the application of CNNs has outperformed other models, particularly when deeper CNNs are used. These deeper networks benefit from balancing parameters and incorporating non-discrimination regularization, which enhances their learning and generalization capabilities. However, it is worth noting that such architectures often come with increased computational complexity and higher data requirements, which may limit their feasibility in low-resource or real-time diagnostic environments. The superior performance of CNNs, especially in handling complex datasets, underscores their importance in future research efforts, provided that considerations around efficiency and scalability are also addressed. Additionally, the application of CNNs has outperformed other models, particularly when deeper CNNs are used. These deeper networks benefit from balancing parameters and incorporating non-discrimination regularization, which enhances their learning and generalization capabilities. However, it is worth noting that such architectures often come with increased computational complexity and higher data requirements, which may limit their feasibility in low-resource or real-time diagnostic environments. The superior performance of CNNs, especially in handling complex datasets, underscores their importance in future research efforts, provided that considerations around efficiency and scalability are also addressed. Overall, future research should focus on integrating fusion models with traditional ML approaches, employing advanced data augmentation techniques, prioritizing dataset diversity to ensure inclusivity across demographics, languages, and cultures, and leveraging the strengths of deeper CNN architectures. These steps will likely address current limitations and pave the way for more accurate and robust models. 4.4. Potential implications These findings have substantial implications for both the educational and healthcare sectors. Integrating AI-based diagnostic tools could revolutionize how dysgraphia is identified and managed, leading to better academic outcomes and support for affected individuals. Furthermore, using these tools can help reduce the stigma associated with dysgraphia by providing clear, objective evidence of the condition. In conclusion, while AI-based models promise to improve dysgraphia detection, ongoing research and development are essential to enhance their accuracy, generalizability, and practical application in real-world educational and healthcare settings. Supporting information This supplementary Excel file contains comprehensive data supporting the systematic review and meta-analysis described in the manuscript. (XLSX) Supplementary data. (DOCX) 23 Sep 2024 Dear Dr. Sajedi, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Nov 07 2024 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols . We look forward to receiving your revised manuscript. Kind regards, Sadiq H. Abdulhussain, Ph.D. Academic Editor PLOS ONE Journal requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf 2. When completing the data availability statement of the submission form, you indicated that you will make your data available on acceptance. We strongly recommend all authors decide on a data sharing plan before acceptance, as the process can be lengthy and hold up publication timelines. Please note that, though access restrictions are acceptable now, your entire data will need to be made freely accessible if your manuscript is accepted for publication. This policy applies to all data except where public deposition would breach compliance with the protocol approved by your research ethics board. If you are unable to adhere to our open data policy, please kindly revise your statement to explain your reasoning and we will seek the editor's input on an exemption. Please be assured that, once you have provided your new statement, the assessment of your exemption will not hold up the peer review process. 3. We note that Figure 11 in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: 1. You may seek permission from the original copyright holder of Figure 11 to publish the content specifically under the CC BY 4.0 license. We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].” 2. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. 4. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information. 5. As required by our policy on Data Availability, please ensure your manuscript or supplementary information includes the following: A numbered table of all studies identified in the literature search, including those that were excluded from the analyses. For every excluded study, the table should list the reason(s) for exclusion. If any of the included studies are unpublished, include a link (URL) to the primary source or detailed information about how the content can be accessed. A table of all data extracted from the primary research sources for the systematic review and/or meta-analysis. The table must include the following information for each study: Name of data extractors and date of data extraction Confirmation that the study was eligible to be included in the review. All data extracted from each study for the reported systematic review and/or meta-analysis that would be needed to replicate your analyses. If data or supporting information were obtained from another source (e.g. correspondence with the author of the original research article), please provide the source of data and dates on which the data/information were obtained by your research group. If applicable for your analysis, a table showing the completed risk of bias and quality/certainty assessments for each study or outcome. Please ensure this is provided for each domain or parameter assessed. For example, if you used the Cochrane risk-of-bias tool for randomized trials, provide answers to each of the signalling questions for each study. If you used GRADE to assess certainty of evidence, provide judgements about each of the quality of evidence factor. This should be provided for each outcome. An explanation of how missing data were handled. This information can be included in the main text, supplementary information, or relevant data repository. Please note that providing these underlying data is a requirement for publication in this journal, and if these data are not provided your manuscript might be rejected. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? Reviewer #1: Partly Reviewer #2: Partly ********** 2. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: N/A ********** 3. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 4. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: No Reviewer #2: Yes ********** Reviewer #1: The Title: AI-Based Models for Dysgraphia Diagnosis on Handwritten Images: A Comprehensive Scoping Review. This paper provides a scoping review to explore available studies employing AI-based models for dysgraphia diagnosis and identify the highest performance models and their challenges to suggest future improvements. The work is good and interesting in its field. However, there are some points need to be considered which are as follows: 1- The authors stated that a scoping review is conducted instead of a systematic review with three mentioned objectives. However, scoping review needs to identify knowledge gaps, explain concepts and investigate research conduct. Are these concepts addressed in this manuscript? The authors should elaborate further to show what has been added to the academic field and what has been discovered and proven. 2- It stated that “All original research articles published until April 2024 …….. were included.”. Is there only one research paper in 2024 submitted in this manuscript? Please review this issue. 3- Some research questions need to be added to the manuscript. This type of review is broad because its purpose is to identify the scope of the literature on the presented topic, so, the research questions that this review can answer should be broad also. 4- Some references should be formatted according to the author's guidelines as there are many references that are not written correctly. 5- There is no need to add [30] in the title of Figure 2 because the information provided is the author’s own. Instead, [30] can be cited in the text of Section 3.1. 6- Figure 3 with the captions of its sub-figures needs more organization. In fact, the titles of most of the tables and figures needs to be reworded to make it more clear to the reader. 7- Ratios listed in figures that containing a circle should be as a percentage like the one in Figure 4. 8- There are many grammatical errors that should be checked and corrected in the entire manuscript. 9- Figure 5 can be presented in another representation and the resolution of Figure 6 and Figure 7 should be increased. 10- The section of the refences should have a title. Reviewer #2: This paper presents a comprehensive scoping review about AI based approaches in literature proposed for Dysgraphia diagnosis from handwritten data. The paper needs major improvements before publishing. Here are my recommendations: 1. The current title, "AI-Based Models for Dysgraphia Diagnosis on Handwritten Images: A Comprehensive Scoping Review," may not accurately reflect the full scope of the paper. While the review covers both offline (image-based) and online handwriting data methods, the title only mentions "handwritten images," which typically refers to offline data. Consider revising the title to encompass both types of data. For example: "AI-Based Models for Dysgraphia Diagnosis Using Online and Offline Handwriting Analysis: A Comprehensive Scoping Review" 2. While your scoping review provides insights into AI-based models for dysgraphia diagnosis, it currently lacks a discussion of existing review/survey papers on this topic. This omission limits the reader's ability to understand how your work builds upon or differs from previous efforts in the field. Include a section that identifies and briefly summarizes existing review papers related to AI-based dysgraphia diagnosis. 3. There are several instances in the paper where specific studies or data points are mentioned without proper citation. For example: "Two of the papers utilized multiple datasets (n = 2, 9%), while only three papers used raw handwritten images (n = 3, 14%)" This statement provides specific numbers and percentages but fails to cite the actual papers being referenced. This oversight occurs in other parts of the manuscript as well. Carefully review the entire manuscript to identify and address all similar instances where specific studies or data points are mentioned without proper citation. 4. What is the need of figure 5. If you want to include it, then please redraw and represent it in a better way. 5. While the acknowledgment of metric heterogeneity and the table of performance indicators are valuable, this crucial aspect warrants deeper analysis. Consider expanding this section to include: brief explanations of each metric and their relevance to dysgraphia diagnosis; trends in metric usage across studies; implications of this heterogeneity for the field's progress; context-specific considerations for dysgraphia evaluation; and potential future directions for standardization. Additionally, discuss how metric choices relate to clinical relevance and practical application. Enhancing this discussion will provide readers with a more comprehensive understanding of the challenges and opportunities in evaluating AI methods for dysgraphia diagnosis, significantly strengthening your scoping review. 6. The review provides a comprehensive overview of AI algorithms used in dysgraphia detection. However, a critical component is missing: an in-depth analysis of the features used in these models. Given that many methods employ traditional machine learning approaches, as mentioned in your review, a thorough examination of input features is essential. This analysis would significantly enhance the review's value and align with your stated goals of identifying effective AI-based models and understanding associated challenges. Add a dedicated section on feature analysis. Categorize and discuss types of features used (e.g., temporal, spatial, pressure-based, kinematic). Analyse trends in feature selection across studies. Examine the relationship between feature choices and model performance. 7. While the current review provides valuable insights into AI-based models for dysgraphia diagnosis, the Discussion and Limitations sections could benefit from deeper analysis and more comprehensive treatment. These sections are crucial for contextualizing findings, critically evaluating the current state of research, and guiding future work. Consider expanding these sections to include: a more nuanced interpretation of the results in light of current clinical practices; a critical comparison of different AI approaches and their relative strengths and weaknesses; a thorough examination of methodological limitations across studies; discussion of potential biases in current research; and analysis of gaps between laboratory findings and real-world application. Minor comments : Improve the presentation language and make sure the manuscript is free from typos ********** what does this mean? ). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org . Please note that Supporting Information files do not need this step. 1 Mar 2025 Note: To enhance clarity, we have highlighted the reviewers' comments and used Arial font for them, while our responses are provided in Times New Roman. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf We have carefully reviewed the PLOS ONE formatting guidelines and ensured that our manuscript adheres to the required style, including file naming conventions. -------------------------------------------------------------------------------------------------------------------------------------------- 2. When completing the data availability statement of the submission form, you indicated that you will make your data available on acceptance. We strongly recommend all authors decide on a data sharing plan before acceptance, as the process can be lengthy and hold up publication timelines. Please note that, though access restrictions are acceptable now, your entire data will need to be made freely accessible if your manuscript is accepted for publication. This policy applies to all data except where public deposition would breach compliance with the protocol approved by your research ethics board. If you are unable to adhere to our open data policy, please kindly revise your statement to explain your reasoning and we will seek the editor's input on an exemption. Please be assured that, once you have provided your new statement, the assessment of your exemption will not hold up the peer review process. The datasets are public. The links for download are available in the data availability section. -------------------------------------------------------------------------------------------------------------------------------------------- 3. We note that Figure 11 in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: 1. You may seek permission from the original copyright holder of Figure 11 to publish the content specifically under the CC BY 4.0 license. We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].” 2. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. We have addressed this issue by removing and replacing the copyrighted image with an open-access image from this source. The new figure adheres to the CC BY 4.0 license requirements. We have also updated the figure caption accordingly. Before: Fig 11 Sample of a typical CNN [67] After: Fig 10 Sample of a typical CNN [67] -------------------------------------------------------------------------------------------------------------------------------------------- 4. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information. We have addressed this by including the captions for all Supporting Information files in a separate file named "DysgraphiaPaper_Files", following the journal’s guidelines. -------------------------------------------------------------------------------------------------------------------------------------------- 5. As required by our policy on Data Availability, please ensure your manuscript or supplementary information includes the following: A numbered table of all studies identified in the literature search, including those that were excluded from the analyses. For every excluded study, the table should list the reason(s) for exclusion. If any of the included studies are unpublished, include a link (URL) to the primary source or detailed information about how the content can be accessed. A table of all data extracted from the primary research sources for the systematic review and/or meta-analysis. The table must include the following information for each study: Name of data extractors and date of data extraction Confirmation that the study was eligible to be included in the review. All data extracted from each study for the reported systematic review and/or meta-analysis that would be needed to replicate your analyses. If data or supporting information were obtained from another source (e.g. correspondence with the author of the original research article), please provide the source of data and dates on which the data/information were obtained by your research group. If applicable for your analysis, a table showing the completed risk of bias and quality/certainty assessments for each study or outcome. Please ensure this is provided for each domain or parameter assessed. For example, if you used the Cochrane risk-of-bias tool for randomized trials, provide answers to each of the signalling questions for each study. If you used GRADE to assess certainty of evidence, provide judgements about each of the quality of evidence factor. This should be provided for each outcome. An explanation of how missing data were handled. We have addressed this by including all required tables and explanations in our supplementary file titled 'Supplementary-Data.' This file contains comprehensive data supporting our systematic review and meta-analysis, organized into nine sheets covering literature search, study selection, and data extraction. It includes a complete list of screened studies, reasons for inclusion/exclusion, keyword lists, search strategies, database-specific extractions, and a detailed breakdown of data extraction responsibilities. Additionally, we provide explanations on quality assessment, data privacy compliance, and accessibility instructions. All figures from the manuscript are included separately in the supplementary files, with proper references provided. For clarity, the detailed contents of 'Supplementary-Data' are also provided below: Supplementary Data Explanation Excel file This supplementary Excel file contains comprehensive data supporting the systematic review and meta-analysis described in the manuscript. The file is organized into nine sheets, each providing specific information related to the literature search, study selection, and data extraction processes employed in our study. Here is a detailed overview of each sheet: Sheet 'Paper': This sheet contains a consolidated list of all studies that were initially screened during the literature search. It includes important information such as study IDs, authors, titles, publication years, and outcomes. This comprehensive listing helps in understanding the scope of literature considered for the review. Sheet 'Included_Paper': This sheet details the studies that were included in the final analysis. It features a color-coded system where entries marked in blue represent data extracted by Yazdan Zandiye Vakili, and entries in white are extracted by Avisa Fallah. This differentiation aids in recognizing the contributions of individual team members to the data extraction process. Sheet 'Keyword': It presents the finalized list of keywords that were used to perform the literature search across various databases. This sheet is crucial for replicating the search methodology used in our study. Sheet 'Source': Contains detailed information about the selected sources for the literature search, including direct links to the search queries applied and specific search strategies used for each source. Sheets 'IEEE_Xplore', 'PubMed', 'Springer', 'Scopus', 'WOS' (World of Science): Each of these sheets provides specific details on the papers extracted from the respective databases. The sheet for each database outlines which papers were considered and includes notes on the extractor, further illustrating the distribution of work among team members. Additional Explanations: Access Instructions: This supplementary material is available directly via contacting the corresponding author. Inclusion and Exclusion Reasons: All the papers, as shown in the PRISMA figure, are categorized based on whether they are review papers, related to the topic, or ultimately included or excluded. Some of the excluded papers were not specifically about dysgraphia itself. Instead, they focused on related topics, such as analyses involving EEGs or brain images, rather than handwritten images. Others were excluded because they were review papers rather than empirical studies. Quality assessment: The quality of the included studies is assessed in the paper and all the explanations can be found in the section 3.4 (Results of Critical Appraisal within Sources of Evidence). Data Protection and Privacy Compliance: All data included in this file have been anonymized and prepared in compliance with applicable data protection and privacy regulations. No personal or sensitive information is disclosed. Version and Date: This document is the final version, created on 20/03/2024, and the final updates were applied on 27/05/2024. For any questions or corrections post-publication, please contact the corresponding author. Use and Navigation Tips: Users are encouraged to review the 'Sheet Guide' included within the file for tips on navigating and understanding the contents effectively. Each sheet is designed to be self-explanatory, with headings and notes providing guidance on the data presented. Figures All the figures in the paper, are included separately in the supplementary files. All of the eleven figures are either created by us or is open accessed and you can see the reference of that in the paper. -------------------------------------------------------------------------------------------------------------------------------------------- Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 4. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer #1: No Reviewer #2: Yes We have revised the manuscript to improve clarity, correctness, and readability. This includes refining the text, enhancing image explanations, and correcting grammatical and scientific errors. The manuscript is now presented in a clear, correct, and unambiguous manner. ________________________________________ 5. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer #1: The Title: AI-Based Models for Dysgraphia Diagnosis on Handwritten Images: A Comprehensive Scoping Review. This paper provides a scoping review to explore available studies employing AI-based models for dysgraphia diagnosis and identify the highest performance models and their challenges to suggest future improvements. The work is good and interesting in its field. However, there are some points need to be considered which are as follows: 1- The authors stated that a scoping review is conducted instead of a systematic review with three mentioned objectives. However, scoping review needs to identify knowledge gaps, explain concepts and investigate research conduct. Are these concepts addressed in this manuscript? The authors should elaborate further to show what has been added to the academic field and what has been discovered and proven. & 3- Some research questions need to be added to the manuscript. This type of review is broad because its purpose is to identify the scope of the literature on the presented topic, so, the research questions that this review can answer should be broad also. We have made several important revisions to ensure the manuscript addresses these concerns effectively: Clarification on Scoping Review Objectives: We have added three new objectives (4, 5, and 6) to the introduction to further elaborate on the contributions of this review to the academic field and highlight the gaps discovered: Objective 4: Examine the integration of machine learning techniques with traditional diagnostic methods. This objective investigates how machine learning techniques are integrated with traditional diagnostic methods for dysgraphia, identifying the benefits and drawbacks of such integrations. It includes an analysis of hybrid models that combine AI with manual assessments or other diagnostic tools. Objective 5: Analyze the training datasets used for developing AI models in dysgraphia diagnosis. This objective analyzes the characteristics, diversity, and adequacy of the training datasets used in AI model development for dysgraphia diagnosis. It addresses the quality of data and its impact on model performance and generalizability. Objective 6: Review technological advancements and their implications for future AI models in dysgraphia diagnosis. This objective reviews recent technological advancements in AI and machine learning, including algorithm advancements, computational hardware, and data collection methods, and discusses their potential impacts on future models for dysgraphia diagnosis. Before: Objectives: 1- Identify available studies that used AI-based models for dysgraphia diagnosis, explore the current dominant use of AI in expediting dysgraphia detection. 2- Explore the challenges of building computational AI-based models under current conditions and the limitations of each study. 3- Determine the best models with high accuracy and performance in dysgraphia diagnosis and examine the effectiveness of these models in dysgraphia detection and possible future improvements in these models. After: Objectives: 1- Identify available studies that used AI-based models for dysgraphia diagnosis and explore the current dominant use of AI in expediting dysgraphia detection. 2- Explore the challenges of building computational AI-based models under current conditions 7 Apr 2025 Dear Dr. Sajedi, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by May 22 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols . We look forward to receiving your revised manuscript. Kind regards, Sadiq H. Abdulhussain, Ph.D. Academic Editor PLOS ONE [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: All comments have been addressed Reviewer #2: (No Response) ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Yes Reviewer #2: Partly ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: I Don't Know Reviewer #2: N/A ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: Most of the comments in the revised version have been properly addressed and no further comments are needed. Reviewer #2: The authors have made commendable efforts to revise the manuscript in response to my previous comments, and several aspects of the paper have indeed improved. However, I believe further refinements are necessary. Some of the earlier comments appear to have been addressed only in a general or superficial manner, lacking the specificity and depth required to significantly enhance the quality. Therefore, additional revision is recommended. In this round of feedback, I have provided more detailed and structured suggestions to help strengthen the manuscript further. Additionally, I have included one new comment (Comment number 5). 1. Regarding your response to my second comment in the first revision: My comment was : "While your scoping review provides insights into AI-based models for dysgraphia diagnosis, it currently lacks a discussion of existing review/survey papers on this topic. This omission limits the reader's ability to understand how your work builds upon or differs from previous efforts in the field. Include a section that identifies and briefly summarizes existing review papers related to AI-based dysgraphia diagnosis." While I appreciate the addition of a discussion on previous review papers in Section 3.2, the response remains too general and does not fully address the concern. The discussion broadly mentions limitations of past reviews without explicitly citing and engaging with specific existing surveys on AI-based dysgraphia diagnosis. For example, one recent and comprehensive review (https://link.springer.com/article/10.1007/s10032-024-00464-z) has examined AI methodologies for dysgraphia diagnosis, and provided a structured analysis of automated systems, yet your manuscript does not acknowledge or differentiate itself from this prior work. Instead, it makes broad claims about the shortcomings of earlier reviews without demonstrating how your study adds new insights beyond what has already been explored. To strengthen the scholarly rigor of your manuscript, I recommend explicitly citing and summarizing existing review papers rather than referring to them in a generalized manner, clearly articulating how your review extends, updates, or differs from prior work, and avoiding generalized critiques of previous reviews unless supported by direct comparisons with specific papers. A more detailed engagement with prior surveys will provide a clearer justification for your review and ensure it is properly positioned within the existing body of literature. Also i suggest changing the title of section 3.2. 2. While the added discussion on evaluation metrics (Section 4.1.2) is a valuable improvement, the section would benefit from further expansion on the clinical and practical implications of metric choices. Specifically, it would be useful to discuss how different metrics align with real-world dysgraphia diagnosis scenarios, considering the trade-offs between false positives and false negatives. Additionally, while the discussion acknowledges metric heterogeneity, it would be strengthened by proposing potential future directions for standardization and best practices in evaluating AI models for dysgraphia. 3. While the inclusion of a dedicated section (3.6) on feature analysis is a valuable addition, I believe there is still an opportunity to provide more depth and specificity. The categorization of features into temporal, spatial, pressure-based, and kinematic is a useful framework, but a more detailed discussion of how these features vary in the context of both online and offline handwriting could add significant value. Given the differences between these two modalities, it would be helpful to explore trends in feature selection across studies focusing on each type of handwriting separately. Additionally, discussing the relationship between feature selection and model performance in more detail—perhaps with specific examples from the reviewed studies—would help clarify how different features contribute to model efficacy. Including relevant citations to support the claims made in this section would also strengthen the discussion, grounding the analysis in the broader body of literature. A brief overview of challenges related to feature standardization across studies would further enrich this section and highlight potential directions for future research. Overall, while the section is a good starting point, expanding on these aspects will provide readers with a more comprehensive understanding of the role that feature selection plays in the development of AI-based dysgraphia diagnostic models. 4.The recommendations outlined in future works section provide a solid foundation for advancing dysgraphia detection models. The focus on combining individual fusion models with traditional machine learning approaches, addressing dataset limitations, and leveraging CNNs for improved performance are highly relevant and aligned with current research trends. However, a few aspects could benefit from further elaboration. For example, while the synergy between fusion models and traditional methods is well-highlighted, additional context or examples on how such integration could be practically implemented would strengthen this point. It would also be useful to address potential challenges, such as computational complexity or data requirements, associated with this integration. Regarding dataset limitations, the mention of data augmentation methods is insightful, but a more detailed explanation of how these techniques can be specifically applied to handwriting data in the context of dysgraphia would be beneficial. Additionally, discussing how future studies could ensure that datasets are more diverse in terms of demographics, language, and cultural context would help provide a broader perspective. Moreover, referencing studies that have successfully employed fusion models, advanced data augmentation, or CNNs in handwriting or related tasks could give a clearer picture of how these methods can be translated into dysgraphia detection. This would also demonstrate that the proposed future work is well-supported by the current body of research, giving it more credibility and a stronger basis for consideration. 5. While Tables 6 to 11 present general advantages and disadvantages of the SVM, KNN, Random Forest algorithms etc..., they would benefit from a more focused discussion contextualized to dysgraphia diagnosis. For example, SVM’s effectiveness with high-dimensional but small handwriting datasets has been reported in [Ref], while KNN’s simplicity makes it a common baseline in early-stage studies despite its sensitivity to noise. Random Forest’s ability to handle mixed feature types (e.g., spatial, temporal, pressure-based) has proven useful in multimodal handwriting analysis [Ref]. Including such application-specific insights would enhance the relevance of these comparisons and help readers understand the trade-offs in algorithm selection for this domain. ********** what does this mean? ). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org . Please note that Supporting Information files do not need this step. 21 May 2025 Third Revision Reviewer #1: Most of the comments in the revised version have been properly addressed and no further comments are needed. Reviewer #2: The authors have made commendable efforts to revise the manuscript in response to my previous comments, and several aspects of the paper have indeed improved. However, I believe further refinements are necessary. Some of the earlier comments appear to have been addressed only in a general or superficial manner, lacking the specificity and depth required to significantly enhance the quality. Therefore, additional revision is recommended. In this round of feedback, I have provided more detailed and structured suggestions to help strengthen the manuscript further. Additionally, I have included one new comment (Comment number 5). 1. Regarding your response to my second comment in the first revision: My comment was : "While your scoping review provides insights into AI-based models for dysgraphia diagnosis, it currently lacks a discussion of existing review/survey papers on this topic. This omission limits the reader's ability to understand how your work builds upon or differs from previous efforts in the field. Include a section that identifies and briefly summarizes existing review papers related to AI-based dysgraphia diagnosis." While I appreciate the addition of a discussion on previous review papers in Section 3.2, the response remains too general and does not fully address the concern. The discussion broadly mentions limitations of past reviews without explicitly citing and engaging with specific existing surveys on AI-based dysgraphia diagnosis. For example, one recent and comprehensive review (https://link.springer.com/article/10.1007/s10032-024-00464-z) has examined AI methodologies for dysgraphia diagnosis, and provided a structured analysis of automated systems, yet your manuscript does not acknowledge or differentiate itself from this prior work. Instead, it makes broad claims about the shortcomings of earlier reviews without demonstrating how your study adds new insights beyond what has already been explored. To strengthen the scholarly rigor of your manuscript, I recommend explicitly citing and summarizing existing review papers rather than referring to them in a generalized manner, clearly articulating how your review extends, updates, or differs from prior work, and avoiding generalized critiques of previous reviews unless supported by direct comparisons with specific papers. A more detailed engagement with prior surveys will provide a clearer justification for your review and ensure it is properly positioned within the existing body of literature. Also i suggest changing the title of section 3.2. Thank you for your insightful comment regarding the clarity of the literature gap our review aims to address. We have revised Section 3.2 to more explicitly highlight how our work fills key gaps left by prior literature. Specifically, unlike earlier reviews that tend to either focus broadly on handwriting difficulties or lean heavily toward clinical perspectives, our review is dedicated exclusively to dysgraphia and places emphasis on artificial intelligence-based diagnostic methods. We address the lack of comprehensive analysis across both traditional machine learning techniques (e.g., SVM, Random Forest) and emerging deep learning architectures (e.g., CNNs). Additionally, we point out that many earlier reviews overlook modern AI approaches or fail to differentiate between dysgraphia and related learning disabilities. By contrast, our paper aims to provide a focused, up-to-date synthesis of AI-based dysgraphia research. To support this clarification, we have incorporated comparisons with existing reviews and added relevant references where appropriate. These revisions enhance the clarity of our paper's unique contribution to the field. -------------------------------------------------------------------------------------------------------------------------------------------- 2. While the added discussion on evaluation metrics (Section 4.1.2) is a valuable improvement, the section would benefit from further expansion on the clinical and practical implications of metric choices. Specifically, it would be useful to discuss how different metrics align with real-world dysgraphia diagnosis scenarios, considering the trade-offs between false positives and false negatives. Additionally, while the discussion acknowledges metric heterogeneity, it would be strengthened by proposing potential future directions for standardization and best practices in evaluating AI models for dysgraphia. For section 4.1.2, we have significantly expanded this section to incorporate the clinical and practical implications of evaluation metrics in the context of dysgraphia diagnosis. We now clearly explain how different metrics align with real-world use cases. For example, we emphasize that recall (sensitivity) is crucial in school-based screening tools, where the primary goal is to avoid missing at-risk children who may need early intervention. Conversely, precision becomes more critical in clinical settings, where minimizing false positives helps prevent unnecessary diagnoses and the stress or resource strain that may follow. We also elaborated on the trade-offs between false negatives and false positives, outlining how they can directly affect intervention timing, educational support, and parental response. To address the challenge of metric heterogeneity across studies, we propose that future research adopt a core set of standardized evaluation metrics (e.g., recall, precision, F1-score, AUC), used consistently across tasks and datasets. We also recommend incorporating public benchmark datasets, k-fold cross-validation, and threshold tuning as standard best practices. Lastly, we underscore the value of interdisciplinary collaboration—including clinicians, educators, and AI researchers—in establishing evaluation standards that are both technically robust and practically meaningful. These additions strengthen the section’s relevance and practical utility. -------------------------------------------------------------------------------------------------------------------------------------------- 3. While the inclusion of a dedicated section (3.6) on feature analysis is a valuable addition, I believe there is still an opportunity to provide more depth and specificity. The categorization of features into temporal, spatial, pressure-based, and kinematic is a useful framework, but a more detailed discussion of how these features vary in the context of both online and offline handwriting could add significant value. Given the differences between these two modalities, it would be helpful to explore trends in feature selection across studies focusing on each type of handwriting separately. Additionally, discussing the relationship between feature selection and model performance in more detail—perhaps with specific examples from the reviewed studies—would help clarify how different features contribute to model efficacy. Including relevant citations to support the claims made in this section would also strengthen the discussion, grounding the analysis in the broader body of literature. A brief overview of challenges related to feature standardization across studies would further enrich this section and highlight potential directions for future research. Overall, while the section is a good starting point, expanding on these aspects will provide readers with a more comprehensive understanding of the role that feature selection plays in the development of AI-based dysgraphia diagnostic models. For your constructive feedback on Section 3.6, we have substantially revised and expanded the section to address each of your suggestions in greater detail. First, we retained the original categorization of features (temporal, spatial, pressure-based, and kinematic) and developed a more detailed comparison of how these feature types differ between online and offline handwriting modalities. For each feature category, we now provide modality-specific explanations that reflect differences in data acquisition, precision, and diagnostic utility. This includes the limitations of inferring temporal and kinematic features from offline data, as well as the enhanced granularity available from digital tablets in online modalities. Second, we have improved subsection (3.6.2) discussing feature selection trends across studies, distinguishing between those focused on online handwriting, offline handwriting, and hybrid approaches. This trend analysis draws from specific studies reviewed in our work, with new in-text examples and citations (e.g., [21], [31], [34], [79]) that illustrate which features were selected and how they impacted model performance. For instance, one study achieved a 15% sensitivity improvement by incorporating kinematic features, while another reported high accuracy using spatial features alone in offline datasets. These examples help clarify the performance implications of different feature types, as you suggested. Third, we have expanded Section 3.6.3 to more thoroughly discuss the relationship between feature selection and model efficacy. We now provide concrete performance outcomes from reviewed studies and explain how certain features align with clinically observable symptoms of dysgraphia (e.g., erratic pressure or prolonged stroke latency), contributing to both diagnostic performance and interpretability. Fourth, as you recommended, we introduced a brief discussion of feature standardization challenges, including variation in tablet sampling rates, device pressure sensitivity, and inconsistent feature definitions across studies. This lack of standardization presents a barrier to reproducibility and comparability across models and was added to highlight future research needs. Finally, we ensured that these revisions are supported by relevant citations to anchor our analysis in the broader literature. These updates strengthen Section 3.6 by providing the specificity, depth, and empirical grounding necessary to help readers better understand the pivotal role of feature selection in AI-based dysgraphia diagnosis. -------------------------------------------------------------------------------------------------------------------------------------------- 4.The recommendations outlined in future works section provide a solid foundation for advancing dysgraphia detection models. The focus on combining individual fusion models with traditional machine learning approaches, addressing dataset limitations, and leveraging CNNs for improved performance are highly relevant and aligned with current research trends. However, a few aspects could benefit from further elaboration. For example, while the synergy between fusion models and traditional methods is well-highlighted, additional context or examples on how such integration could be practically implemented would strengthen this point. It would also be useful to address potential challenges, such as computational complexity or data requirements, associated with this integration. Regarding dataset limitations, the mention of data augmentation methods is insightful, but a more detailed explanation of how these techniques can be specifically applied to handwriting data in the context of dysgraphia would be beneficial. Additionally, discussing how future studies could ensure that datasets are more diverse in terms of demographics, language, and cultural context would help provide a broader perspective. Moreover, referencing studies that have successfully employed fusion models, advanced data augmentation, or CNNs in handwriting or related tasks could give a clearer picture of how these methods can be translated into dysgraphia detection. This would also demonstrate that the proposed future work is well-supported by the current body of research, giving it more credibility and a stronger basis for consideration. We greatly appreciate your thoughtful feedback on the future works section and the encouragement to provide further elaboration. In response, we have revised the section to strengthen the depth and clarity of our recommendations, addressing all key points raised. To elaborate on the integration of fusion models with traditional machine learning, we now describe how handcrafted or statistical features, extracted from traditional preprocessing pipelines (e.g., kinematic or pressure-based analysis), can be combined with deep feature representations and passed into conventional classifiers such as SVMs or Random Forests. This hybrid approach enables the integration of domain knowledge with learned representations, resulting in more robust and interpretable models. However, we also acknowledge potential challenges, such as increased computational complexity, limited scalability, and the difficulty of optimally weighting heterogeneous feature sources. These trade-offs are now explicitly discussed to give a more balanced and practical outlook. Regarding dataset limitations and augmentation, we expanded the discussion to clarify how techniques such as shearing, shifting, mirroring, and flipping can simulate natural variations in handwriting, including stroke rhythm, pressure, and alignment—all of which are known indicators of dysgraphia. This specificity helps ground the relevance of data augmentation strategies to dysgraphia-related motor and spatial irregularities. Additionally, we have addressed the need for greater dataset diversity by emphasizing the importance of including samples from diverse demographics, cultural contexts, and languages. We also expanded on dataset diversity via multi-center collaborations and non-Latin scripts (e.g., Arabic, Devanagari), ensuring inclusivity and reducing bias. This broader inclusion is crucial for enhancing generalizability and mitigating algorithmic bias. To support these future directions, we have incorporated citations to studies that have successfully implemented fusion strategies, augmentation techniques, and CNN architectures in handwriting and related tasks. These references demonstrate the feasibility and prior success of the methods discussed, reinforcing the credibility of the proposed directions. -------------------------------------------------------------------------------------------------------------------------------------------- 5. While Tables 6 to 11 present general advantages and disadvantages of the SVM, KNN, Random Forest algorithms etc..., they would benefit from a more focused discussion contextualized to dysgraphia diagnosis. For example, SVM’s effectiveness with high-dimensional but small handwriting datasets has been reported in [Ref], while KNN’s simplicity makes it a common baseline in early-stage studies despite its sensitivity to noise. Random Forest’s ability to handle mixed feature types (e.g., spatial, temporal, pressure-based) has proven useful in multimodal handwriting analysis [Ref]. Including such application-specific insights would enhance the relevance of these comparisons and help readers understand the trade-offs in algorithm selection for this domain. Thank you for your valuable feedback. To fully address Comment 5, we have revised Tables 6–11 to include dysgraphia-specific insights in the advantages and disadvantages of each algorithm (SVM, KNN, Random Forest, K-means, AdaBoost, CNN), focusing on their application to handwriting data and dysgraphia diagnosis challenges. Specifically, we: Updated Table Titles: Changed the titles of Tables 6–11 to explicitly reference dysgraphia diagnosis (e.g., Table 6: “Advantages and Disadvantages of SVM Algorithm in Dysgraphia Diagnosis”) to clarify the context. Added Dysgraphia-Specific Insights: Incorporated application-specific advantages and disadvantages, such as SVM’s effectiveness with small, high-dimensional handwriting datasets, KNN’s role as a baseline despite noise sensitivity, Random Forest’s ability to handle multimodal features, K-means’ clustering of dysgraphic patterns, AdaBoost’s enhancement of weak classifiers, and CNN’s strength in complex image patterns despite dataset limitations. Included Trade-off Discussion: Added a new paragraph at the end of Section 3.5 to discuss trade-offs in algorithm selection, highlighting how dataset characteristics ( Submitted filename: RESPON~1.PDF 15 Jun 2025 Dear Dr. Sajedi, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jul 30 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols . We look forward to receiving your revised manuscript. Kind regards, Sadiq H. Abdulhussain, Ph.D. Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #2: (No Response) ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #2: Yes ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #2: N/A ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #2: Yes ********** Reviewer #2: I appreciate your thorough revisions and the improvements made throughout the manuscript. Most of my earlier comments have been addressed satisfactorily. However, one key issue remains regarding the treatment of related literature in Section 3.2. In your response, you mention prior reviews in a general sense, but you do not cite or engage directly with the recent and highly relevant review by Kunhoth et al. (2024), "Automated systems for diagnosis of dysgraphia in children: a survey and novel framework". This work overlaps closely in scope with your own and offers a comprehensive review of AI-based dysgraphia diagnosis, including both traditional and deep learning approaches, feature-level discussions, and even non-ML tools. I strongly encourage you to cite this work explicitly and include a short structured comparison that clarifies how your review builds upon or differs from it. This will improve the scholarly positioning and rigor of your manuscript. Once this addition is made, I believe the manuscript will be in strong shape for publication. ********** what does this mean? ). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org 3 Jul 2025 Note: To enhance clarity, we have highlighted the reviewers' comments and used Arial font for them, while our responses are provided in Times New Roman. Third Revision Reviewer #2: I appreciate your thorough revisions and the improvements made throughout the manuscript. Most of my earlier comments have been addressed satisfactorily. However, one key issue remains regarding the treatment of related literature in Section 3.2. In your response, you mention prior reviews in a general sense, but you do not cite or engage directly with the recent and highly relevant review by Kunhoth et al. (2024), "Automated systems for diagnosis of dysgraphia in children: a survey and novel framework". This work overlaps closely in scope with your own and offers a comprehensive review of AI-based dysgraphia diagnosis, including both traditional and deep learning approaches, feature-level discussions, and even non-ML tools. I strongly encourage you to cite this work explicitly and include a short structured comparison that clarifies how your review builds upon or differs from it. This will improve the scholarly positioning and rigor of your manuscript. Once this addition is made, I believe the manuscript will be in strong shape for publication. Our response: Thank you for highlighting the importance of comparing our work with other comprehensive reviews. In response, we have expanded the discussion in the related work section to include a detailed comparison with the recent review by Kunhoth et al., 2024 [1], which offers valuable insights into learning disorders and predictive methods, both machine learning and non-machine learning based. We clarify that while Kunhoth et al. provide a broad overview and effective categorization across various disorders, our review focuses specifically on dysgraphia and delivers more in-depth technical analysis, including a detailed breakdown of individual models, feature selection, and implementation strategies. We emphasize how our review contributes beyond general knowledge by offering model-level insights, highlighting recent AI advancements, and synthesizing results from studies involving modern, hybrid approaches like the integration of spatial and dynamic handwriting features. Before: “The publication dates span nearly a decade, from 2015 to the present (Fig 3). Of the 21 selected papers, 8 (n = 8, 38%) [34,35,38,42-44,47] are conference papers, and the remaining 13 (n = 13, 62%) [20,31-33,36,39,40,41,45,46,48-57] are journal articles. The papers originate from various countries, including India, Qatar, Spain, Switzerland, Slovakia, the Czech Republic, Israel, France, the USA, Malaysia, Morocco, and Austria. Additionally, some papers are collaborative efforts involving researchers from multiple nationalities. In addition, only nine of the included studies are open-access (n = 9, 42%) [20,31-33,40,41,45,49], and the 12 other papers are private (n = 12, 58%) [34-39,42-44,46-48]. While our review provides detailed insights into the current AI methodologies for diagnosing dysgraphia, it is essential to contextualize this within the broader academic dialogue. Previous reviews have provided foundational knowledge in this field; however, most adopt a clinical perspective, prioritizing diagnostic criteria and behavioral assessments over AI’s potential to transform dysgraphia diagnosis [50,51]. Moreover, few are dedicated solely to dysgraphia, often embedding it within broader discussions of specific learning disabilities (SLDs) like dyslexia and dyscalculia, thus diluting focus on dysgraphia-specific challenges [52-54]. Even reviews exploring AI applications, such as [55,56], typically emphasize traditional machine learning methods (e.g., SVM, AdaBoost) and overlook state-of-the-art approaches, notably the integration of traditional and modern techniques, such as combining handwriting image analysis with dynamic features like pen pressure or motor patterns, which our review examines in detail. Additionally, many prior reviews are outdated, failing to reflect recent AI advancements. Our work addresses these gaps by providing a dysgraphia-specific, up-to-date synthesis of AI applications, comprehensively covering both traditional and cutting-edge methods. This contribution highlights the multidisciplinary relevance of AI in dysgraphia diagnosis through handwritten image analysis, underscoring its potential impact across educational and healthcare settings. Despite growing interest, a recent slowdown in publications suggests a bottleneck in data availability, emphasizing the need for improved data collection and sharing to sustain research momentum in this critical area.” After: “The publication dates span nearly a decade, from 2015 to the present (Fig 3). Of the 21 selected papers, 8 (n = 8, 38%) [34,35,38,42-44,47] are conference papers, and the remaining 13 (n = 13, 62%) [20,31-33,36,39,40,41,45,46,48-58] are journal articles. The papers originate from various countries, including India, Qatar, Spain, Switzerland, Slovakia, the Czech Republic, Israel, France, the USA, Malaysia, Morocco, and Austria. Additionally, some papers are collaborative efforts involving researchers from multiple nationalities. In addition, only nine of the included studies are open-access (n = 9, 42%) [20,31-33,40,41,45,49], and the 12 other papers are private (n = 12, 58%) [34-39,42-44,46-48]. While our review provides detailed insights into the current AI methodologies for diagnosing dysgraphia, it is essential to contextualize this within the broader academic dialogue. Previous reviews have provided foundational knowledge in this field; however, most adopt a clinical perspective, prioritizing diagnostic criteria and behavioral assessments over AI’s potential to transform dysgraphia diagnosis [50,51]. Moreover, few are dedicated solely to dysgraphia, often embedding it within broader discussions of specific learning disabilities (SLDs) like dyslexia and dyscalculia, thus diluting focus on dysgraphia-specific challenges [52-54]. Even reviews exploring AI applications, such as [55,56], typically emphasize traditional machine learning methods (e.g., SVM, AdaBoost) and overlook state-of-the-art approaches, notably the integration of traditional and modern techniques, such as combining handwriting image analysis with dynamic features like pen pressure or motor patterns, which our review examines in detail. Additionally, many prior reviews are outdated, failing to reflect recent AI advancements. One of the more comprehensive recent works is the review by Kunhoth et al. (2024) [57], which provides a broad overview of various learning disorders and explores both machine learning and non-machine learning approaches for their prediction. While their work offers valuable general insights and effectively groups and compares different models, it lacks the depth of analysis found in our study. In contrast, our review includes more specific technical details, considers a wider array of models, and delves into how they are implemented. It also provides deeper guidance and recommendations, making it more actionable for researchers and practitioners. Together, these additions make our review not only complementary to previous works but also a necessary update for the current state of AI research in this domain. Our work addresses these gaps by providing a dysgraphia-specific, up-to-date synthesis of AI applications, comprehensively covering both traditional and cutting-edge methods. This contribution highlights the multidisciplinary relevance of AI in dysgraphia diagnosis through handwritten image analysis, underscoring its potential impact across educational and healthcare settings. Despite growing interest, a recent slowdown in publications suggests a bottleneck in data availability, emphasizing the need for improved data collection and sharing to sustain research momentum in this critical area.” Other Minor Changes: As part of the future recommendations, we have added an example [2] from our recently published work to demonstrate the practical implementation of integrating fusion models with traditional approaches. This real-world case enhances the clarity and applicability of the proposed direction, offering readers a concrete understanding of how such integration can be effectively realized. References: 1. Kunhoth, J., Al-Maadeed, S., Kunhoth, S., Akbari, Y., & Saleh, M. (2024). Automated systems for diagnosis of dysgraphia in children: a survey and novel framework. International Journal on Document Analysis and Recognition (IJDAR), 27(4), 707-735. 2. Vakili, Y. Z., Fallah, A., Esmaeili, K., & Mirfazeli, F. S. (2025). Revolutionizing Dysgraphia Detection: Combining Feature Fusion with Non-Discriminatory Regularization. 2025 11th International Conference on Web Research (ICWR), 250–256. https://doi.org/10.1109/ICWR65219.2025.11006233 Submitted filename: Response to Reviewers.pdf 7 Jul 2025 AI-Driven Approaches for Dysgraphia Diagnosis Using Online and Offline Handwriting Data: A Comprehensive Scoping Review PONE-D-24-28058R3 Dear Dr. Sajedi, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. If you have any questions relating to publication charges, please contact our Author Billing department directly at authorbilling@plos.org. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Sadiq H. Abdulhussain, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #2: All comments have been addressed ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #2: Yes ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #2: N/A ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #2: Yes ********** Reviewer #2: All comments have been addressed. ********** what does this mean? ). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #2: No ********** PONE-D-24-28058R3 PLOS ONE Dear Dr. Sajedi, I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now being handed over to our production team. At this stage, our production department will prepare your paper for publication. This includes ensuring the following: * All references, tables, and figures are properly cited * All relevant supporting information is included in the manuscript submission, * There are no issues that prevent the paper from being properly typeset You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps. Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing. If we can help with anything else, please email us at customercare@plos.org. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Sadiq H. Abdulhussain Academic Editor PLOS ONE