As artificial intelligence continues to transform modern ophthalmology, MadhuNetrAI has emerged as a significant indigenous innovation for large-scale screening of diabetic retinopathy (DR) in India. Developed through a collaborative effort between the Dr Rajendra Prasad Centre for Ophthalmic Sciences, the Union Health Ministry’s e-Health division, and Wadhwani AI, the platform represents one of the first fully validated, India-specific AI tools designed for automated detection and grading of diabetic eye disease.
AI-Driven Screening for Rapid Detection
MadhuNetrAI analyses 45-degree macula-centred colour fundus photographs captured using either tabletop or handheld retinal cameras, including low-cost devices commonly used in community screening programmes.
Within seconds, the system processes these images and classifies them as referable or non-referable diabetic retinopathy. As a result, healthcare providers can rapidly triage patients and ensure timely referrals, even in healthcare settings where ophthalmologists are not immediately available.
Robust Training Using Diverse Datasets
The development team trained the algorithm using a combination of international open-source datasets and carefully curated retinal images from AIIMS Delhi. This approach enabled the model to learn from diverse data sources while maintaining relevance to the Indian population.
Consequently, the system achieved strong performance during initial testing and validation.
High Accuracy Demonstrated in Clinical Validation
During retrospective validation conducted at AIIMS Delhi, MadhuNetrAI demonstrated 93 percent sensitivity and 95 percent specificity in detecting diabetic retinopathy.
Furthermore, community-based validation using low-cost retinal cameras produced even stronger results, achieving 96 percent sensitivity and 97 percent specificity. These findings confirmed the tool’s effectiveness in real-world community screening settings, where access to advanced imaging equipment may be limited.
Prospective Evaluation Confirms Reliability
Subsequently, researchers conducted a prospective evaluation between October 2023 and February 2025 across AIIMS Delhi and NCR vision centres, analysing 2,507 retinal images.
The results showed 98 percent sensitivity and 96 percent specificity, along with a Kappa score of 0.92, indicating near-perfect agreement with expert retinal graders. These findings demonstrate the system’s high reliability and strong clinical concordance with ophthalmology specialists.
Advancing Equitable Eye Care Through Collaboration
MadhuNetrAI illustrates how academic institutions, public health agencies, and technology organisations can collaborate to create a scalable, context-appropriate AI solution for improving eye care access in India. By enabling early detection of diabetic retinopathy at the community level, the platform has the potential to reduce preventable vision loss and strengthen public health screening programmes.
Recognition at the India AI Impact Summit
Highlighting its national significance, the RPC team and Wadhwani AI presented the MadhuNetrAI model at the India AI Impact Summit held at Bharat Mandapam. The innovation was recognised among the top 10 AI models, reflecting its potential to transform diabetic retinopathy screening and digital health delivery in India.