Vision-driven diagnostic intelligence through deep learning for retinal health analytics

Abstract

The objective of this study is to employ retinal fundus images as a vascular health monitoring tool in the classification of Age-related Macular Degeneration (ARMD) and Diabetic Retinopathy (DR), as well as lesions that are Exudates (EX), Microaneurysms (MCA), hemorrhages (HE), and Drusen (DN). One model was developed for the classification of DR and ARMD in this project, utilizing the ResNet101 architecture. The precision archive for DR is 97%, while the precision archive for ARMD is 81%. Yet another model for classifying lesions in retinal fundus images was implemented by employing the ResNet101 architecture. Lesion models, EX, MCA, HE, and DN, were archived with 83%, 91%, 90%, and 87% precisions. U-net was used to segment each lesion after it was detected. The lesion segmentation model was achieved with an accuracy of 98%, 99%, 97%, and 96%. Our research highlights the promise of automated systems for vascular health monitoring using pretrained deep learning models.