Eyesight: Integrative AI For Non-Invasive Diabetic Retinopathy Detection using Pupillometry and Ensemble Deep Learning

The disorder Diabetic Retinopathy causes severe vision impairment and blindness predominantly among diabetic people. Early detection serves as the crucial factor to stop severe outcomes of diseases. The research introduces EyeSight which serves as an advanced noninvasive technology to detect and classify DR through assessment of pupillometry data using deep learning models. EyeSight conducts analysis on pupillary light reflex (PLR) to gauge diabetic retinopathy (DR) severity through five diagnostic levels ranging from Healthy to Mild DR to Moderate DR to Proliferative DR and culminating in Severe DR. Three deep learning models namely ResNet and DenseNet and EfficientNet analyze pupillometric data to establish its classifications. An ensemble model combines individual predictions from other models in order to enhance classification accuracy rates. EyeSight operates from Streamlit to provide healthcare professionals with easy options for uploading pupillometry measurements and instant DR severity diagnoses supported by confidence ratings. The system functions as an affordable noninvasive solution that speeds up diagnosis time while preventing permanent vision loss among diabetic patients.