Detection and Classification of Multiple Sclerosis from Brain MRIs by Using MobileNet 2D-CNN Architecture

Abstract

Deep Learning based object detection and classification has been widely investigated for the neuroimaging. The Magnetic Resonance Imaging (MRI) data serve as diagnostic tool for detection and classification of brain disorders like Parkinson, Alzheimer’s disease (AD) and Multiple Sclerosis (MS). Further, use of Convolutional Neural Network (CNN) framework helps in developing predictive models from the available MRI images. The aim of this work is to develop a CNN based model with pre-trained MobileNet model to detect and classify the Multiple Sclerosis using MRI image dataset. In this paper, We have proposed a pretrained MobileNet-2D-CNN architecture for accurate prediction of multiple sclerosis from various MRI images. Initially, the proposed model extract the images from MRI images of affected patient with MS and healthy control. We have used the MRI images to train the MobileNet - 2D-CNN model for identification of MS features map that predict the MS. The proposed architecture has been validated on the standard MRI scans. We have also performed a class activation map for the interpretation of prediction as provided by the proposed model which represents the behavior of neurons at the early stages. The proposed approach achieves the classification accuracy of 98.15% and AUC = 1.00.