Deep Learning Framework for OCT Based Cataract Detection

Abstract

Cataract is a leading cause of visual impairment worldwide, commonly associated with symptoms such as blurred vision, glare sensitivity, reduced contrast perception, and gradual vision loss, highlighting the need for early awareness and accurate diagnosis. This study proposes an integrated deep learning based framework for cataract detection, segmentation, and risk prediction using Optical Coherence Tomography (OCT) eye images and patient lifestyle data. Initially, OCT images undergo preprocessing to enhance image quality and reduce noise. A median filter is employed to suppress speckle and impulse noise, followed by a Gaussian filter to achieve smoothness while preserving essential ocular structures. Contrast enhancement and Region of Interest (ROI) extraction are performed to localize cataract-affected regions.

For automated cataract region analysis, two deep learning algorithms U-Net and a Deep Convolutional Neural Network (DCNN) are implemented. U-Net is used for precise pixel level segmentation of cataract regions, while DCNN is applied for feature extraction and classification of cataract severity. Comparative evaluation demonstrates that U-Net achieves superior segmentation accuracy, whereas DCNN provides reliable classification performance. Additionally, a numerical dataset incorporating food habits, average daily alcohol intake, and non-alcohol consumption is utilized for lifestyle-based risk prediction. The results indicate a higher cataract risk among individuals with unhealthy dietary patterns and regular alcohol intake. Overall, the proposed framework supports early detection, risk stratification, and preventive ophthalmic care.