Exploring Image Transforms derived from Eye Gaze Variables for Progressive Autism Diagnosis
Abigail Copiaco 1 , Christian Ritz 2 , Yassine Himeur 1 , Valsamma Eapen 3 , Ammar Albanna 4 and Wathiq Mansoor 1 · 2025
Read Exploring Image Transforms derived from Eye Gaze Variables for Progressive Autism Diagnosis by Abigail Copiaco 1 , Christian Ritz 2 , Yassine Himeur 1 , Valsamma Eapen 3 , Ammar Albanna 4 and Wathiq Mansoor 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-09065.txt.
Introduction
Abigail Copiaco and five co-authors wrote a paper in 2025 on an artificial intelligence tool designed to speed up the diagnosis of autism spectrum disorder in children. Autism affects communication, behavior, and focus. Current diagnostic techniques are effective, but they are time-intensive and carry high social and economic costs.
The system uses transfer learning, a method that applies knowledge learned from earlier data to a new task. It converts variables derived from eye gaze into image transforms that the system can analyze. The authors argue that the approach could allow periodic diagnosis at home, with less stress for individuals and caregivers. Privacy is preserved because gaze data are transformed into images rather than stored as raw recordings. The method could also improve communication between guardians and therapists, with regular updates on progress and changing support needs.
Catalog record
| CREATOR | Abigail Copiaco 1 , Christian Ritz 2 , Yassine Himeur 1 , Valsamma Eapen 3 , Ammar Albanna 4 and Wathiq Mansoor 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-09065
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Exploring Image Transforms derived from Eye Gaze Variables for Progressive Autism Diagnosis Abigail Copiaco 1 , Christian Ritz 2 , Yassine Himeur 1 , Valsamma Eapen 3 , Ammar Albanna 4 and Wathiq Mansoor 1 Abstract The prevalence of Autism Spectrum Disorder (ASD) has surged rapidly over the past decade, posing significant challenges in communication, behavior, and focus for affected individuals. Current diagnostic techniques, though effective, are time-intensive, leading to high social and economic costs. This work introduces an AI-powered assistive technology designed to streamline ASD diagnosis and management, enhancing convenience for individuals with ASD and efficiency for caregivers and therapists. The system integrates transfer learning with image transforms derived from eye gaze variables to diagnose ASD. This facilitates and opens opportunities for in-home periodical diagnosis, reducing stress for individuals and caregivers, while also preserving user privacy through the use of image transforms. The accessibility of the proposed method also offers opportunities for improved communication between guardians and therapists, ensuring regular updates on progress and evolving support needs. Overall, the approach proposed in this work ensures timely, accessible diagnosis while protecting the subjects’ privacy, improving outcomes for individuals with ASD. I Introduction The number of individuals affected by Autism Spectrum Disorder (ASD) has increased rapidly over the past decade, with a 175% rise in diagnosed cases during this period [ 1 ] . ASD is primarily characterized by challenges in communication, attention, and behavior [ 2 ] . While the severity of these challenges varies between individuals, the specialized care required often leads to substantial social and economic costs. Current diagnostic techniques, including gold standard methods such as the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview-Revised (ADI-R), though effective, are often time-intensive, resource-demanding, and reliant on highly trained professionals [ 3 , 4 ] . These limitations create bottlenecks in early diagnosis, delaying critical interventions and contributing to disparities in access to care. Moreover, while recent advances in neuroimaging, genetics, and proteomics have uncovered potential biomarkers for ASD, their clinical applicability remains limited due to challenges such as high costs, dependence on specialized equipment, and lack of robust validation across diverse populations. Additionally, while behavioral traits such as reduced eye contact and atypical fixation patterns have been identified as significant markers for ASD [ 5 ] , existing approaches to quantify and analyze these markers often rely on manual annotation or simplistic computational models that fail to fully capture the complexity of gaze dynamics. Another significant gap lies in the lack of effective integration of image-based analysis with eye-tracking data [ 6 ] . While eye-tracking technology has shown promise in identifying ASD-related gaze patterns, there is limited research leveraging advanced machine learning models to classify these patterns automatically [ 7 , 8 ] . Furthermore, the potential of image transformations—such as heatmaps, scan paths, and fixation maps—to enhance the representation of gaze data remains underexplored [ 9 ] . These transformations can uncover spatial, temporal, and frequency-based patterns that are crucial for accurate diagnosis but have yet to be systematically applied in ASD research [ 10 ] . Aside from this, implementing transforms to images anonymizes data and protects the privacy of the individuals concerned. In this study, we address these gaps by proposing a transfer learning-based approach to classify visual attention patterns in children with ASD using the GoogleNet architecture. The method also evaluates Typically Developing (TD) individuals to establish a comparative baseline. Our focus is on assessing the diagnostic performance of three distinct input representations—heatmaps, scan paths, and fixation maps—derived from eye-tracking data. By analyzing key eye gaze variables, such as reduced eye contact and atypical fixation patterns, this work aims to identify early markers of ASD that are essential for timely diagnosis and intervention while protecting their privacy. The primary contributions of this research are summarized as follows: • Classify visual attention patterns in children with ASD and typically developing (TD) individuals using machine learning methods. • Compare the effectiveness of heat maps, scan paths, and fixation maps as input representations for detecting ASD-related visual attention anomalies, while ensuring privacy by using anonymized gaze data instead of raw facial or video recordings. • Identify early markers of ASD, such as reduced eye contact and atypical fixation patterns, to improve early diagnosis strategies. • Optimize the performance of the model through fine-tuning for each input type to achieve high classification accuracy. II Background ASD is a complex neurodevelopmental condition characterized by challenges in social interaction, communication, and repetitive behaviors. Early and accurate diagnosis is crucial for providing timely interventions that can significantly improve outcomes [ 11 , 12 ] . II-A Current Trends in Autism Diagnosis Traditional diagnostic frameworks, such as the Autism Diagnostic Observation Schedule (ADOS) [ 3 ] and the Autism Diagnostic Interview-Revised (ADI-R) [ 4 ] , remain gold standards but are resource-intensive and require specialized training. To address these limitations, researchers have proposed the use of more accessible screening tools that enable remote developmental assessments, as well as wearable devices that monitor atypical behaviors and physiological responses. These innovations provide real-time, objective data that complement traditional methods. Firstly, genetic studies have identified numerous genes linked to ASD, while neuroimaging techniques like MRI have revealed structural and functional brain abnormalities [ 13 , 14 ] . Furthermore, proteomic research has identified several possible biomarkers in biological samples [ 15 ] , offering promising potential for ASD diagnosis. However, despite these advancements, extensive validation across diverse populations and cases remains essential. Several studies also challenge these claims, arguing that autism is a behavioral disorder rather than a neurological condition, as current brain imaging techniques cannot reliably detect it [ 16 ] . II-B Artificial Intelligence for Autism Diagnosis Current research trends promote the potential of Artificial Intelligence (AI) in transforming autism diagnosis by enabling the analysis of large datasets and uncovering subtle patterns [ 17 , 18 ] . Machine learning algorithms may also be used to train on behavioral, genetic, and neuroimaging data for ASD classification. AI-based video analysis detects atypical facial expressions, gestures, and speech patterns, while natural language processing (NLP) identifies linguistic anomalies in speech development [ 19 ] . Predictive models are being developed to assess ASD risk in infants and toddlers [ 20 ] , facilitating earlier interventions. Several challenges persist in the deployment of these innovations. For example, ensuring data privacy and addressing biases in datasets are critical for ethical implementation [ 21 ] . Access to advanced diagnostic technologies is often limited in low-resource settings, which can lead to disparities in care [ 22 ] . Furthermore, many emerging tools require rigorous validation and standardization before integration into clinical practice [ 23 , 24 ] . To address these limitations, this research introduces the innovative use of eye gaze variables as a key feature for detecting autism in children. By focusing on these behavioral markers, the approach aims to provide a non-invasive and effective diagnostic tool. Additionally, the integration of image transformation techniques serves a dual purpose: enhancing data privacy by hiding sensitive information and uncovering underlying patterns and features that may remain hidden in the raw inputs. This approach not only protects user confidentiality but also enhances the model’s ability to extract meaningful insights,