Eye Sight: Integrative AI for Non-Invasive Diabetic Retinopathy Detection using Pupillometry and Ensemble Deep Learning

Diabetic Retinopathy (DR) is the leading cause of blindness and vision impairment among people with diabetes and timely diagnosis is important to prevent severe consequences. Traditional DR screening methods are invasive and expensive collection techniques. Eyesight devises an innovative methodology to address this challenge by exploiting pupillometry, a noninvasive technique which measures the response of the pupil to light stimulus. Eye Sight combines this with ensemble deep learning models (ResNet, DenseNet, and EfficientNet) to classify DR severity across five stages: Mild DR, Moderate DR, Proliferative DR, and Severe DR. Early DR is detected by the system in an available, cheap, and accurate method. Eyesight is implemented on the Streamlit platform and provides clinicians the ability to upload pupillometry data to receive real time results with high confidence thus enabling timely intervention. Integrating pupillometry with advanced machine learning models improves the diagnostic efficiency and scalability and creates global access to early DR detection and treatment.