How harshahemanthkumar/sih2026-project works
Retina-AI is a MATLAB-based diabetic-retinopathy screening and telemedicine system. A technician captures a fundus image; the documented workflow checks image quality, classifies DR severity, analyzes lesions and retinal structures, produces Grad-CAM explanations, and generates a report with referral triage. Separate branches cover model training and external validation, plus Simulink district-scale capacity simulation. The README documents these stages and their intended relationships, but no source excerpts were supplied, so implementation-level call ownership and wiring cannot be verified; directory paths identify unsampled subsystems.
Main entry points: MATLAB-Scripts DR Grading · dr_resnet50_model.mat ResNet-50 Model · TRAIN_ALL_DATASETS_MODEL.m Model Training · PART9_EXTERNAL_VALIDATION_MESSIDOR_IDRID.m External Validation
Components 21
Screening workflow
- Image Quality Gate
MATLAB-Scripts - Recapture Feedback
MATLAB-Scripts - Image Enhancement
MATLAB-Scripts
AI analysis
- DR Grading
MATLAB-Scripts - Lesion Segmentation
MATLAB-Scripts - Retinal Structure Analysis
MATLAB-Scripts - Grad-CAM Explainability
MATLAB-Scripts
Clinical output
- Clinical Report Generation
MATLAB-Scripts - PDF Reports
reports
Model lifecycle
- ResNet-50 Model
models/dr_resnet50_model.mat - U-Net Model
models/unet_exudate_model.mat - Model Training
MATLAB-Scripts/TRAIN_ALL_DATASETS_MODEL.m - External Validation
MATLAB-Scripts/PART9_EXTERNAL_VALIDATION_MESSIDOR_IDRID.m
Network simulation
- Screening Throughput Simulation
simulink/telemedicine_screening_workflow.slx - M/M/c Queue Model
simulink/telemedicine_screening_workflow.slx
Other components
- PHC Technician
- Fundus Camera
- Ophthalmologist
- Training Datasets
- MESSIDOR / IDRiD
- Rural PHC Network
Connections 24
- PHC Technician → Fundus Camera: captures image
- Fundus Camera → Image Quality Gate: submits image
- Image Quality Gate → Image Enhancement: enhances borderline
- Image Quality Gate → DR Grading: passes adequate
- Image Quality Gate → Recapture Feedback: rejects image
- Recapture Feedback → PHC Technician: requests recapture
- Image Enhancement → DR Grading: passes image
- DR Grading → ResNet-50 Model: infers with
- DR Grading → Lesion Segmentation: provides image
- Lesion Segmentation → U-Net Model: segments with
- Lesion Segmentation → Retinal Structure Analysis: analyzes structures
- DR Grading → Grad-CAM Explainability: explains grading
- DR Grading → Clinical Report Generation: provides grade
- Lesion Segmentation → Clinical Report Generation: provides lesion evidence
- Retinal Structure Analysis → Clinical Report Generation: provides anatomy
- Grad-CAM Explainability → Clinical Report Generation: provides heatmaps
- Clinical Report Generation → PDF Reports: generates PDF
- Clinical Report Generation → Ophthalmologist: refers urgent cases
- Training Datasets → Model Training: supplies samples
- Model Training → ResNet-50 Model: trains model
- MESSIDOR / IDRiD → External Validation: supplies images
- External Validation → ResNet-50 Model: evaluates model
- Rural PHC Network → Screening Throughput Simulation: defines workload
- Screening Throughput Simulation → M/M/c Queue Model: simulates throughput