Deep Learning-based Analysis of Brain MRI for Enhanced Diagnosis of Multiple Sclerosis

A neuro degenerative disease that worsens over time, Multiple Sclerosis (MS), where early and exact diagnosis of the disease will lead to an improved treatment. In this research, a new method is introduced to diagnose MS through deep learning techniques by developing and assessing an enhanced neural network model called MS-CNN. Here, we propose MS-CNN that utilizes Deep Convolution Neural Networks (DCNNs) to automatically analyze brain MR images for improved diagnostic accuracy and efficiency. In particular we provided an end-to-end evaluation of MS-CNN on a dataset made up by MRI images divided in healthy-MS. Evaluation on CIFAR-100 dataset was carried out against two popular existing models, ResNet-50 and VGG-16 with MS-CNN. The parameters used for evaluation include sensitivity, specificity, accuracy, precision, and F1 score implemented and done with best results followed by AUC-ROC and processing time. MS-CNN does much better across the board among the models and get higher accuracy up to 94.0%, sensitivity up to 94.0% and specificity up to 95.0%. These results show that MS-CNN not only increases accuracy for diagnosing the disease, but also substantially decreases processing time to provide a feasible solution when instantaneous detection and monitoring are needed with regards to such cases in real-time possible by deep learning technology.