IIT Madras AI Diagnostic Model Achieves Clinical-Grade Accuracy
A research team from the Department of Computer Science and Engineering at IIT Madras has published validation results for a deep learning model capable of diagnosing diabetic retinopathy — the leading preventable cause of blindness in India — with an accuracy of 96.3 per cent in a clinical validation cohort of 48,000 retinal images drawn from four ophthalmology centres across Tamil Nadu and Telangana.
The model, named DR-Scan Net, was trained on a dataset of 2.4 million labelled retinal fundus images, making it one of the largest training corpora assembled for this specific diagnostic task in an Indian clinical context. The team deliberately weighted the training dataset toward Indian eye morphologies and diabetic complication profiles, which differ in certain statistically significant ways from the Western datasets on which earlier leading models were predominantly trained.
Designed for the Last Mile
The most distinctive engineering decision in the project is the model's optimisation for deployment on affordable smartphone-based retinal cameras rather than the high-cost fundus photography equipment typical of tertiary care hospitals. The researchers partnered with a Chennai-based medical device startup to adapt the model for a slit-lamp attachment that connects to standard Android phones, reducing the hardware cost for a screening station from approximately ₹3.5 lakh to ₹18,000.
This design choice directly addresses India's diabetic retinopathy screening challenge. Of India's estimated 77 million people living with diabetes, only a small fraction have access to regular specialist ophthalmological examinations. The addressable screening gap is largest in rural and semi-urban settings where specialist eye care is absent, and where general practitioners and community health workers with limited ophthalmology training could theoretically operate the simplified device if the diagnostic AI is reliable enough to reduce specialist referral bottlenecks.
Clinical Integration Pathway
The IIT Madras team has submitted results to the Central Drugs Standard Control Organisation for medical device software classification, the regulatory pathway applicable to AI-based diagnostic tools under the Indian Medical Devices Rules. Regulatory clearance is expected within 12 to 18 months. The team has also entered into a knowledge transfer agreement with the medical device startup, which will commercialize the integrated hardware-software screening product.
The Indian Council of Medical Research has expressed interest in incorporating the validated model into its national diabetic eye disease screening protocol, which is currently deployed at approximately 600 district hospitals. Ophthalmologists have cautioned that AI screening results must be verified by a specialist before treatment decisions are made, and that the model's real-world performance in diverse deployment conditions requires post-market surveillance before its role in clinical pathways is definitively established.
Abhijit Chowdhury
Staff Reporter
Editorial administrator for Eastern Times.
India's Semiconductor Mission Approves Three Chip Design Centres in Bengaluru
Submit a Perspective for editorial consideration at editorial@easterntimes.in. All submissions are moderated for professional credentials and civil exchange.