CataLight: A Novel Lightweight Approach for Cataract Severity Detection Using Digital Imaging on Edge Devices

Cataract is a leading cause of blindness around the world, and many people who are affected reside in rural areas with limited access to health services. The traditional approach to cataract detection is a manual examination by ophthalmologists. Currently, no healthcare system globally can ultimately support the detection of cataract severity from digital eye images, particularly in places with limited access to healthcare. Early cataract detection can help people to improve their vision and quality of life. The study proposes CataLight, a lightweight and efficient method for detecting cataract severity classified as mild, severe, or normal on edge devices using digital camera eye images. Further, CataLight is designed to function offline and is suitable for deployment in regions without internet access. The approach uses advanced computer vision and image processing techniques to evaluate cataract severity and processes images from standard digital cameras, such as those found in smartphones. Before the classification, a lightweight segmentation model extracts the lens region by generating an iris mask. In the study, MobileNetV2, NASNetMobile, DenseNet121, and custom Convolutional Neural Network (CNN) models are tried out for cataract severity detection, and MobileNetV2 achieves the highest accuracy of 99.52%. By providing an easy-to-use, cost-effective diagnostic tool, CataLight aims to improve early detection of cataracts, ultimately reducing preventable blindness and enhancing healthcare accessibility in underserved communities.