Interpretable Retinal Disease Classification from OCT Images Using Deep Neural Network and Explainable AI

Deep Learning Models (DNN) are being used extensively for medical image classification such as MRI, OCT, x-ray in recent years. The proposed model revolves around the analysis of macular Optical Coherence Tomography (OCT) images to distinguish three eye-related anomalies: Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), and accumulation of Drusen from OCT images of patients. At first, the dataset was acquired and various pre-processing steps were performed on it. Then we performed a split on the dataset into train-test-validation sets with different numbers of images in each of them. Afterward, we applied pre-trained Resnet, Inception V3, and EfficientNet models in order to classify the images. From our experiment, we achieved the best accuracy of 96.9% from ResNet. Finally, we applied Explainable AI (XAI) framework though the LIME framework in an attempt to explain the reasons for misclassifications. Alongside achieving slightly better accuracy compared to the base model, the purpose of our research is to explain the reasons behind classification errors which can be utilized in the future to develop better models.