Enhancing Ophthalmic Diagnostics: CNNs in Cataract Detection
Kartik Solanki, Shaurya Singh, Aman Yadav, Vikrant Sharma, Shashank Awasthi, Satvik Vats · 2024
Cataracts have been one of the most prevalent eye disorders, which often can cause significant visual impairment due to the clouding of the eye's lens. This condition can worsen in future, often leading to severe vision problems and even blindness. Consequently, detecting cataracts is paramount in mitigating the associated risks and preventing the onset of blindness. Throughout the years there has been quite good progress in leveraging cutting-edge technology, especially due to machine learning, to improve the perfection and efficacy of cataract detection. Convolutional Neural Networks (CNNs) have made an appearance as a commanding implement for computerized the systematization of eye images in the context of cataract identification. Proposed research was more concerned with fine-tune the process of cataract identification, aiming to increase the accuracy while minimizing the data loss. To accomplish this, we conducted a series of experiments, with a key focus on manipulating a critical parameter: the number of training epochs. The proposed research revealed a compelling relationship between the number of training epochs and the accuracy and loss of data in CNN. As we delved into a spectrum of epoch values, a clear pattern emerged: the higher the number of epochs, the more refined and potent the model became. In this research, a significant milestone was reached when utilizing a generous number of 90 epochs, resulting in an impressive accuracy rate of 97.37%.