Detection and Classification of Eye Diseases using Hybrid Deep Features with Decision Tree Algorithm

In ophthalmic diagnostics, it is essential to identify certain ocular conditions in fundus images. This work presents the Decision Tree (DT) algorithm, an automated multi-label deep learning classification system that integrates deep feature representations from pairs of fundus images (e.g., left and right eyes) to diagnose multiple ocular disorders. The benchmark structured fundus image dataset Ocular Disease Intelligent Recognition (ODIR-5K) has been used to classify fundus images into normal, diabetes, glaucoma, cataract, age-related macular degeneration, hypertension, pathological myopia, and others. Our suggested approach consists of three steps: pre-processing, deep feature extraction based on transfer learning concepts, and classification using a classical decision tree classifier. The study initiates a comprehensive image pre-processing procedure, including circular border cropping, image resizing, data balancing, and data augmentation. The work aims to categorize a dataset of eye diseases using DT and Hybrid Features (HF) based transfer learning models. This approach has been applied to identify the eye disease dataset based on fused characteristics from the EffiientB6 and Densnet169 models as transfer learning models. The final classifier (Decision Tree) achieved an F1-score of 99.02%, accuracy of 99.02%, precision of 99.04%, and recall of 99.01%.