Deep Learning-Based Eye Disease Classification Using Convolutional Neural Networks

The major contributors to vision impairment include three eye diseases, namely glaucoma and cataracts and diabetic retinopathy (DR). The standard diagnosis techniques use extended time periods while depending on testers’ subjective interpretation and skilled vision doctors for results. This research introduces CNNs for developing an automatic system which classifies eye diseases through deep learning methodology. Data enhancement techniques, including contrast adjustment and normalization followed by resizing and data augmentation, transformed the fundus image dataset to enhance general model performance. The CNN structure employs various convolutional layers together with Rectified Linear Unit activation combined with batch normalization and maximum pooling accompanied by dropout for regularization. The model used categorical cross-entropy loss under the Adam optimizer to achieve its results, which were evaluated through precision, F1score, accuracy, recall, and AUC-ROC metrics. The CNN model was conducted with 91.5% accuracy while maintaining AUC scores of 0.90 or higher across all examined disease categories, thus validating its high discriminating power. The study results from the confusion matrix analysis demonstrated that small errors in the model mostly occurred between glaucoma and cataract identification. The model shows strong performance but continues to face issues stemming from imbalanced classes together with inconsistent image qualities.