NEW TRANSFER LEARNING-BASED HYBRID DEEP MODEL FOR MULTICLASS EYE DISEASE CLASSIFICATION WITH INCEPTION V3 AND TAILORED CNN BLOCKS

Glaucoma, diabetic retinopathy, and cataract-each of these eye ailments are threatening to the complete loss of vision if not detected early.To aid this, a hybrid deep learning model was built by an amalgamation of InceptionV3 along with the additional custom convolution layers and dense layers for enhanced classification of retinal fundus images.The model was trained on a well-balanced dataset of 4,200 images after resizing to 224 224 3 for efficiency, with 80% of the data going to training, 10% to validation, and 10% to testing sets.Preprocessing implemented noise addition, contrast enhancement, and normalization.Training was performed for 50 epochs with adaptive learning rate scheduling, achieving accuracies of 99.49% on training data, 95.84% on validation data, and 96.32% on testing data.The good results emphasize the applicability of this model in clinical practice.