Advanced Machine Learning Approaches for Cataract Diagnosis Using CNN

Among the most common diseases affecting eyes and impairing vision, cataract is one of the most significant, and the best way to decrease risks and prevent blindness is to detect cataracts properly and early; the activation and loss functions were modified to train the network with smaller kernels, fewer training parameters and layers: This cataract detection based on traditional algorithms with a feature representation model leads to false classification, either positive or negative, mainly due to its dependence on the classification process done by an ophthalmologist. The activation and loss functions were modified to train the network with smaller kernels, fewer training parameters and layers This research study aims to enhance the accuracy value and minimize data loss in the cataract identification process by performing an experiment wherein the manipulation process was carried out through epoch switching. Accordingly, as compared to other pre-trained CNN models, Cataract Net reduces average running time and lowers highly computational cost. For optimization of the proposed network, we utilized an Adam optimizer. For model training, we collect and enhance 1130 images of fundus—both cataract and non-cataract—to 4746 images. By the year 2025, cataract-induced blindness cases may as high as 40 million. Augmentation is made to the dataset before training a model to avoid the problem of over- fitting. Experimental results show that this proposed method performs better compared to existing most advanced techniques for cataract detection with an average accuracy of 99.13.