Intelligent retinal screening, designed for earlier clinical attention.
Upload a retinal fundus image to run the AI-assisted screening prototype: 5-class diabetic retinopathy scoring, confidence, a Grad-CAM explainability overlay, and referral guidance for clinician review.
- 01CaptureUpload a retinal fundus photograph (JPG/PNG).
- 02Quality CheckHeuristic checks for resolution, illumination, and focus.
- 03AI AnalyzePrototype classifier scores the image across 5 DR severity classes.
- 04ResultsConfidence, probability distribution, and explainability overlay.
- 05Clinician ReviewConfirm, request recapture, or refer — session-local only.
Research prototype
About Drishti AI
A complete screening workflow
Drishti AI is an AI-assisted retinal image screening prototype designed to support earlier clinical attention for diabetic retinopathy, especially where access to specialist review is limited. Upload a fundus image, check its quality, review five-class predictions and Grad-CAM explainability, then route the result to a clinician.
Important note
This is a screening-support aid, not a diagnostic system or medical device. Predictions and visualizations must not be used as the sole basis for diagnosis or treatment. Qualified clinical oversight is always required.
Drishti AI is an AI-assisted diabetic retinopathy screening support tool. This build implements the full upload → quality check → analysis → results → clinician review workflow end to end. Because no trained EfficientNet checkpoint or Grad-CAM implementation exists in the connected repository, analysis currently runs through an isolated, clearly-labeled mock heuristic classifier (see lib/mock-inference.ts) instead of a real model. Every API response and UI surface that uses it is tagged accordingly. This tool does not replace an ophthalmologist and does not provide treatment recommendations.