Using Machine Learning to Detect Different Eye Diseases from OCT Images

Abstract

Diseases or damage to the retina that cause adverse effects are one of the most common reasons people lose their sight at an early age. Today, machine learning techniques, which give high accuracy results in a short time, have been used for disease detection in the biomedical field. Optical coherence tomography, an advanced ophthalmic imaging technique to display the cross-section of retinal layers, is one of the important tools used for the determination, analysis and treatment design of retinal diseases. The aim of this study is to detect eight retinal diseases that can occur in the eye and cause permanent damage as a result, using machine learning from eye tomography images. For this purpose, hyperparameter settings were applied to six deep learning models, training was performed on the OCT-C8 dataset and performance analyzes were made. The performance of these hyperparameter-tuned models was also compared with previous eye disease detection studies in the literature, and it was seen that the classification success of the hyperparameter-tuned DenseNet121 model presented in this study was higher than the success of the other models discussed. The fine-tuned DenseNet121 classifier achieved 97.79% accuracy, 97.69% sensitivity, and 97.79% precision for the OCT-C8 dataset.