Reti-Cardio: An AI-Driven Multi model Oculo-mics System for Retinal Disease Classification and Cardiovascular Risk Profiling

M. NAGAKEERTHI1, B. SRUTHI2

1Assistant professor, 2MCA Final semester, Master of Computer Applications, Sanketika Vidya Parishad Engineering College Vishakhapatnam, Andhra Pradesh, India.

Sruthiborra1234@gmail.com

ABSTRACT
Cardiovascular disease (CVD) remains the leading cause of global mortality, with conventional diagnostic methods being invasive, costly, and fundamentally unsuitable for population-scale early screening. The retina offers a unique non-invasive window into systemic vascular health, with morphological changes in retinal microvasculature — including arteriole-to-venule ratio (AVR), vessel tortuosity, bifurcation angle, and fractal dimension — serving as clinically validated biomarkers for hypertension, coronary artery disease, diabetic retinopathy, glaucoma, and age-related macular degeneration. Despite substantial evidence establishing these oculo-mics associations, no unified research framework currently exists that simultaneously performs quantitative retinal vascular morphometry and multi-class ocular disease classification within a single, explainable, and clinically interpretable system.

This paper proposes Reti Cardio, a multimodal oculo-mics framework integrating classical computer-vision-based vascular morphometry with transfer-learning-based deep classification for simultaneous cardiovascular risk profiling and ocular disease detection from a single retinal fundus photograph. A dual-pathway architecture is implemented: a classical image-processing pipeline extracts and quantifies four vascular morphometric parameters mapped to established clinical thresholds, while an EfficientNet-B0 model with two-phase transfer learning performs five-class ocular disease classification (AMD, Diabetic Retinopathy, Glaucoma, Hypertensive Retinopathy, and Normal) on a consolidated dataset of approximately 3,400 retinal fundus images drawn from RFM id, MAPLES-DR, STARE, and ORIGA. The system achieves a final validation accuracy of 93.1% and a system AUROC of 0.941, with Gradient-weighted Class Activation Mapping (Grad-CAM) confirming anatomically appropriate model attention. A Framingham-inspired cardiovascular risk scoring module, a live AUROC performance tracking mechanism, and a rule-based Oculo-mics AI Diagnostic Chatbot are integrated within a Flask-based web interface, together establishing Reti-Cardio as a novel, deployable contribution to population-scale cardiovascular and ophthalmic screening.

Keywords:

Oculo-mics, Retinal Fundus Analysis, EfficientNet-B0, Transfer Learning, Arteriole-to-Venule Ratio, Grad-CAM, Explainable AI, Cardiovascular Risk Profiling, Framingham Risk Score, Diabetic Retinopathy, Glaucoma, AMD, Hypertensive Retinopathy.