anand000009999/Ocunexa-Clinical-Kiosk

OCUNEXA — Explainable AI for Diabetic Retinopathy Screening in Rural India (SIH26038)

An explainable, human-in-the-loop AI screening and clinical decision-support system designed to enable accessible, reliable and scalable Diabetic Retinopathy (DR) screening in rural and resource-constrained healthcare settings.

🩺 Overview

Diabetic Retinopathy is a major cause of preventable vision loss, yet access to regular retinal screening remains limited in rural areas due to shortage of ophthalmologists, large screening populations, variable-quality fundus images and poor connectivity.

OCUNEXA (LaxmiAI / NetraX) addresses these challenges through an intelligent retinal screening pipeline that combines automated image quality assessment, retinal structure analysis, lesion detection (YOLO26 Nano), vascular graph analysis (GAT GNN), explainable AI (KAN B-Splines) and confidence-based ophthalmologist escalation.

The system is designed to support healthcare workers at Primary Health Centres (PHCs) while keeping ophthalmologists in the clinical decision loop.

💡 Proposed Solution

The complete screening pipeline follows:

Portable Fundus Camera → Image Quality Assessment → Enhancement → Retinal Structure & ETDRS Mapping → YOLO26 + GNN/GAT Analysis → Feature Fusion → 5-Level DR Grading → Confidence Gate → Explainable Report → Referral / Monitoring

🏗️ System Architecture

                 PORTABLE FUNDUS CAMERA
                           │
                           ▼
                IMAGE QUALITY ASSESSMENT
             (Tenengrad, Illumination, FOV)
                           │
                 ┌─────────┴─────────┐
                 ▼                   ▼
             [Accept]            [Reject] ──► Recapture Required
                 │
                 ▼
            IMAGE ENHANCEMENT
       (CLAHE, Green Channel, Denoise)
                 │
                 ▼
     RETINAL STRUCTURE & ETDRS MAPPING
     (Optic Disc, Fovea, Grid Zones)
                 │
      ┌──────────┼──────────┐
      ▼          ▼          ▼
    YOLO26      GNN        KAN
   Segment    Vascular    Fusion
      │          │          │
      └──────────┼──────────┘
                 │
                 ▼
           FEATURE FUSION
                 │
                 ▼
          5-LEVEL DR GRADING
                 │
                 ▼
           CONFIDENCE GATE
            ┌────┴────┐
            ▼         ▼
     High Confidence  Low / Conflicting
            │         │
            ▼         ▼
      Report /      Ophthalmologist
      Referral         Review