Paper Details
Abstract
Cataracts, defined as the clouding of the eye’s lens leading to impaired vision, represented a significant global health concern. Early and accurate detection is essential for preventing disease progression and improving patients' quality of life. Traditional cataract detection and grading methods relied heavily on the expertise of ophthalmologists, which can be costly and inaccessible to certain demographic groups seeking timely care. This study introduced a computer-assisted diagnostic approach for detecting and classifying cataracts using fundus retinal images. The proposed method utilized multiple pretrained neural networks, including Inception and Xception, to extract features from fundus images. These features were then classified using CNNs, ANNs, and MLPs into four severity levels: normal, mild, moderate, and severe. A PCA-based feature fusion mechanism further enhanced prediction accuracy. Evaluated on a hybrid dataset of 1600 fundus images annotated by an expert ophthalmologist, our proposed method achieved 98.74% accuracy, outperforming existing approaches, offering a reliable, affordable, and accessible tool for early cataract detection, contributing significantly to global eye health.