AI-Powered Retinal Screening
Detect Diabetic
Retinopathy.
Before It's Too Late.
Upload a fundus photograph and receive 5-stage diabetic retinopathy grading, vessel segmentation maps, and explainable AI confidence scores within seconds.
6
Public Datasets
3
Research Papers
Validated on datasets from
Worry-free retinal screening.
Clinical-grade insights from a single fundus photograph.
5-Stage DR Classification
Cascade architecture with confidence-based escape hatch. No DR → Mild → Moderate → Severe → PDR with 87% overall accuracy.
Vessel Segmentation Map
Multi-scale Attention U-Net achieves Dice 0.891, trained on DRIVE, STARE, CHASE_DB1, HRF, and FIVES — surpassing prior SOTA by 3.7–7.7%.
Explainable AI Overlays
Grad-CAM, Grad-CAM++ and Score-CAM overlays reveal exactly which retinal regions drive each prediction. Full transparency.
Image Quality Gating
10-metric DREAM-RFI framework rejects low-quality images before analysis — unique among retinal AI tools, ensuring reliable outputs.
How ChakshurakhsakAI works.
Three steps from image to insight.
01
Upload Fundus Image
Drag & drop or browse. Quality is auto-assessed via the DREAM-RFI 10-metric framework before analysis begins.
02
AI Analysis Pipeline
ConvNeXt Large backbone processes RGB + FADHE-enhanced green channel through a 4-stage cascade classifier with confidence routing.
03
Explainable Results
Receive DR grade, confidence score, vessel segmentation map, and Grad-CAM overlay. Every prediction is interpretable.
No DR Found
Recommend annual screening.
DR Detected
Refer for ophthalmology review. Severity graded.
Severe / PDR
Urgent ophthalmologist referral required.
Low Image Quality
Recapture with better lighting.