iHealth DR is a privacy‑preserving SaaS that screens for Diabetic Retinopathy (DR) from fundus images. It combines a Vision Transformer (RETFound‑based) backbone with explainability (Grad‑CAM & LIME) and clinician‑ready PDF reports.
ViT initialized with RETFound weights, fine‑tuned for DR severity classification. Robust augmentation, calibration, and thresholding.
Grad‑CAM for attention maps; LIME for local feature attribution. Outputs are embedded in the report and UI for clinician review.
Users (patients or technicians) capture right & left eye images. The system returns prediction, heatmaps, and guidance. Supports asynchronous rendering of LIME with placeholders.
On‑device encryption for sensitive data, no cloud upload by default, role‑based access, and audit logs. Federated learning is planned for multi‑site personalization.
