Detection of Multiple Sclerosis Using Convolutional Neural Networks: A Comparative Study

Brain imaging is one of the most essential methods in detecting brain lesions, which can be caused by many diseases. Multiple sclerosis (MS) is an autoimmune disease that attacks the body's nervous system, including the brain and spinal cord. MS diagnosis is carried out normally by differential diagnosis because most neuro diseases exhibit the same symptoms, so it is carried out by eliminating similar diseases, which may cause false diagnosis. In this research, we intend to diagnose MS from magnetic resonance images using machine and deep learning methods by classifying the images into MS or healthy subjects. Many approaches were tested, including convolutional neural network (CNN), random forest, and support vector machine. The best results were obtained from CNN implemented with VGG16 architecture, with the following results (accuracy: 98.44%; specificity: 99.4%; sensitivity: 97.56
^%
). This deep learning approach proved to be efficient in MS diagnosis and would help clinicians to reassure their proposed diagnosis.