AI-Driven Diabetic Retinopathy Screening: Multicentric Validation of AIDRSS in India
Amit Kr Dey, Pradeep Walia, Girish Somvanshi, Abrar Ali, Sagarnil Das, Pallabi Paul, Minakhi Ghosh · 2025
Read AI-Driven Diabetic Retinopathy Screening: Multicentric Validation of AIDRSS in India by Amit Kr Dey, Pradeep Walia, Girish Somvanshi, Abrar Ali, Sagarnil Das, Pallabi Paul, Minakhi Ghosh (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-2501-05826.txt.
Introduction
Dey and co-authors tested an AI screening system for diabetes-related eye damage in India, in a study published in 2025. The disease can blind, and India has too few eye specialists, above all in rural areas. The system reads photos of the back of the eye and flags the damage automatically. The team ran a multi-center check in Kolkata with 5,029 people. Over ten thousand eye photos were graded both by the system and by standard means.
The system's deep learning core was trained on a large set of images, with an image-sharpening step before grading. The study measured how well its calls matched expert judgment and counted how much disease it found. A screening tool that holds up across centers could reach patients who now go unchecked. The authors present the work as evidence for scaling automated screening in places where specialists are scarce.
Catalog record
| CREATOR | Amit Kr Dey, Pradeep Walia, Girish Somvanshi, Abrar Ali, Sagarnil Das, Pallabi Paul, Minakhi Ghosh |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-05826
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
AI-Driven Diabetic Retinopathy Screening: Multicentric Validation of AIDRSS in India Amit Kr Dey, Pradeep Walia, Girish Somvanshi, Abrar Ali, Sagarnil Das, Pallabi Paul, Minakhi Ghosh Abstract Purpose: Diabetic retinopathy (DR) is a major cause of vision loss, particularly in India, where access to retina specialists is limited in rural areas. This study aims to evaluate the Artificial Intelligence-based Diabetic Retinopathy Screening System (AIDRSS) for DR detection and prevalence assessment, addressing the growing need for scalable, automated screening solutions in resource-limited settings. Approach: A multicentric, cross-sectional study was conducted in Kolkata, India, involving 5,029 participants and 10,058 macula-centric retinal fundus images. The AIDRSS employed a deep learning algorithm with 50 million trainable parameters, integrated with Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing for enhanced image quality. DR was graded using the International Clinical Diabetic Retinopathy (ICDR) Scale, categorizing disease into five stages (DR0 to DR4). Statistical metrics including sensitivity, specificity, and prevalence rates were evaluated against expert retina specialist assessments. Results: The prevalence of DR in the general population was 13.7%, rising to 38.2% among individuals with elevated random blood glucose levels. The AIDRSS achieved an overall sensitivity of 92%, specificity of 88%, and 100% sensitivity for detecting referable DR (DR3 and DR4). These results demonstrate the system’s robust performance in accurately identifying and grading DR in a diverse population. Conclusions: AIDRSS provides a reliable, scalable solution for early DR detection in resource-constrained environments. Its integration of advanced AI techniques ensures high diagnostic accuracy, with potential to significantly reduce the burden of diabetes-related vision loss in underserved regions. keywords Artificial Intelligence, diabetic retinopathy screening, deep learning, neural networks, Contrast Limited Adaptive Histogram Equalization (CLAHE) 1 Introduction Diabetic Retinopathy (DR) [ 1 ] [ 2 ] is a serious microvascular complication of diabetes mellitus and a significant global health concern. It is the leading cause of preventable vision loss among working-age adults worldwide, reflecting the growing burden of diabetes in both developed and developing countries. The gradual progression of DR, often without early symptoms, necessitates systematic screening to enable timely intervention and prevent irreversible vision impairment. Early detection is crucial to halt the progression of DR, which otherwise may result in costly treatment options or permanent blindness. 1.1 Significance India is facing a dual challenge with the increasing prevalence of diabetes and limited access to specialized ophthalmic care, particularly in rural and underserved regions [ 3 ] . According to a nationwide screening program conducted by the All-India Ophthalmological Society (AIOS) in 2014, 21.7% of individuals with diabetes were found to have DR. This highlights the substantial public health challenge posed by DR and the need for targeted interventions. Despite this, awareness about diabetes-related complications and access to regular retina screening remain inadequate, leaving a large segment of the population at risk of vision-threatening stages of DR. The significance of early detection cannot be overstated. Routine retina screening facilitates the timely identification of early-stage DR, which can be managed effectively to prevent progression to advanced stages. Key measures for addressing DR include: • Promoting Awareness : Educating the public on maintaining optimal glycemic control and the necessity of regular eye examinations. • Routine Screening : Encouraging annual fundus examinations, even for asymptomatic individuals. • Timely Referrals : Ensuring that treating physicians recognize and prioritize ophthalmic evaluations for diabetic patients. 1.2 Research Objectives This study evaluates the effectiveness of the Artificial Intelligence-based Diabetic Retinopathy Screening System (AIDRSS) developed by ARTELUS™, an automated deep learning algorithm for DR detection. The primary objectives of this research are: 1. To assess the performance of AIDRSS in classifying retinal fundus images based on the International Clinical Diabetic Retinopathy (ICDR) Scale. 2. To enhance the accuracy of the AIDRSS algorithm by incorporating advanced image preprocessing techniques, such as Contrast Limited Adaptive Histogram Equalization (CLAHE). 3. To determine the prevalence of DR across multiple screening centers in Kolkata, West Bengal, India, and examine its correlation with random blood glucose (RBG) levels and other clinical parameters. 1.3 Implications of the Study This research has the potential to significantly influence public health policy and clinical practice. The key implications include: • Scalability and Accessibility : Demonstrating how AI-based systems like AIDRSS can address the scarcity of retina specialists, especially in resource-limited settings. • Clinical Accuracy : Validating the reliability of AIDRSS through rigorous evaluation against gold-standard retina specialist assessments. • Policy Development : Informing healthcare strategies for integrating AI technologies into national DR screening programs, thereby reducing the burden of diabetes-related blindness. By bridging the gap between technology and clinical application, this study highlights the transformative potential of automated screening systems in addressing the growing burden of DR in India and similar healthcare contexts globally. 2 Materials and Methods 2.1 Study Design This was a retrospective study using data collected from routine screenings at multiple centers in Kolkata, India. The data comprised retinal fundus images collected between January and March 2023. A total of 5,029 participants were screened, and their de-identified retinal images were analyzed retrospectively. The validation was performed on a withheld dataset from a separate site, ensuring an independent evaluation of AIDRSS. 2.2 Validation Methodology We validated the AIDRSS system’s performance on a completely withheld dataset from an independent screening site, comprising 1,200 images not included in the training or development datasets. Additionally, five-fold cross-validation was employed to assess the model’s generalizability across diverse patient subsets. Each fold reserved data from a unique center, ensuring no overlap between training and testing images. The withheld site validation tested the system’s robustness when applied to a new population, while cross-validation provided insight into its consistency across different subsets of the study population. Performance metrics, including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated for both the cross-validation and withheld datasets. To ensure statistical robustness, confidence intervals (95% CI) for all performance metrics were computed using bootstrap resampling with 1,000 iterations. This method provides a non-parametric estimate of the variability in the calculated metrics, offering a reliable measure of the system’s performance. Additionally, McNemar’s test was performed to compare the performance of the AIDRSS against baseline or other models on paired data. These statistical approaches ensure a thorough and accurate assessment of the model’s diagnostic capabilities. 2.3 Participants The study screened a total of 5,029 adults during the specified study period. Among this population, a significant proportion, 4,261 individuals (84.7%), were unaware of their diabetes status. Furthermore, 4,692 participants (93.3%) were uninformed about the potential microvascular complications associated with diabetes, including the risk of permanent blindness resulting from uncontrolled diabetes. 2.4 Diabetic Retinopathy Screening System (AIDRSS) The AIDRSS utilized in this study was the Artelus Fundus camera (Fig 1), a portable and easy-to-operate non-mydriatic automatic device for capturing retinal fundus photographs. The camera significantly reduced the time required to capture fundus images and was complemented by the AIDRSS software (DRISTI A.I.) [ 4 ] , an Artificial Intelligence-based program designed to analyze fundus images. The AIDRSS utilized a deep learning algorithm with a modified neural network architecture comprising approximately 50 million trainable parameters and 250 layers deep. The network parameters were learned using the Adam optimizer algorithm. Figure 1: Artelus Fundus Camera and DR prevalence in general population. 2.5 CLAHE as an image processing technique To enhance