Deep learning and classic machine learning models in the automatic diagnosis of multiple sclerosis using retinal vessels

Abstract

This study aims to automatically detect multiple sclerosis (MS) in terms of the changes in retinal vessels using Scanning laser ophthalmoscopy (SLO) images. Although much research has been done to diagnose MS patients, these diagnostic techniques have always been based on using Magnetic resonance imaging (MRI) images which cannot be a complete technique in diagnosing this disease. Using SLO images and examining the condition of its vessels using computer technology, biomarkers in the vessel can be identified to help diagnose MS patients. However, in the first step, the color images are converted to gray and after that are improved using a combination of algorithm Tylor Coye and DWT, then, the images are segmented and retinal vessels are extracted. Besides, two different techniques are used in classification stage. In the first technique, classic Machine learning different features are extracted from the resulting regions and entered into several multiple classifiers, the results of which give us an accuracy of 72, moreover in the second technique segmented images enter the transfer learning model and ultimately lead us to 98 accuracy in the distinction between MS patients and Healthy Controls (HCs).