Implementation of a Deep Learning Algorithm for The Detection of Cataract Disease Severity in Eyes

Cataracts cloud the lens of the eye, resulting in blindness and poor vision. Approximately 18 million people are blind due to cataracts, according to the WHO. Prompt diagnosis is essential to avoid problems. This research creates a deep learning model that uses Convolutional Neural Networks (CNN) to categorise cataract severity into four groups: hypermature, normal, immature, and mature. Transfer learning is used with three CNN architectures: VGG16, VGG19, and ResNet50. Experiments from various eras were carried out using a labelled eye picture dataset for training. Using the confusion matrix, performance was assessed by calculating F1-score, recall, accuracy, and precision. As a result, VGG19 had an accuracy of 88% and F1-score of 0.87, while VGG16 had the best accuracy. ResNet50 had the lowest model accuracy, achieving 63% accuracy and F1-score of 0.59. These results highlight how important CNN selection is when using deep learning to diagnose cataracts.