Diagnosis of retinal diseases using the vision transformer model based on optical coherence tomography images

Abstract

Deep learning is rapidly becoming the state of the art, leading to enhanced performance in various medical diagnosis. In recent years, convolutional neural network (CNN) has been used to diagnose retinal disease and has proven its superiority in detection and classification tasks. Vision transformer is a new image classification model that has been proposed in 2020. It does not rely on any CNN and completely performs based on the transformer structure which has a different feature extraction method from CNN. In this study, diagnosis of retinal disease using vision transformer was presented using optical coherence tomography (OCT) images. A multi-class classification layer in the vision transformer model was used to group the OCT images into the normal and three abnormal type, Choroidal Neovascularization (CNV), Drusen, and Diabetic Macular Edema (DME). The proposed method achieved a accuracy of 95.76%, sensitivity of 95.77% and specificity of 98.59% in detecting CNV, DME and DRUSEN. Results showed that the classification accuracy of vision transformer is higher than that of other traditional CNN models. The performance of vision transformer was evaluated with different performance metrics like accuracy, sensitivity, and specificity, which proved that vision transformer is a statistically significant method than other standard CNN architectures in classifying retinal diseases using OCT images. This technology enables early diagnosis of retinal diseases, which may be useful for optimal treatment to reduce vision loss.