Processing of brain images has some difficulties because of the large data size and complexity of the data. Deep learning facilitates hierarchicical feature extraction automatically. However the optimization of deep nets and validation of extracted features is critical in neuroimage processing. In multiple sclerosis, detection of the lesion is quite important for diagnosis, treatment, and follow up. Changes in brain morphology and white matter lesions are most significant findings in MS, where this diagnose and follow up is done nowadays by experts in the field subjectively. In this study, 40 MS patients scanned twice with an interval of 6 months, earning 80 MR images, which are grouped into 2 and tagged as having an MS lesion or not, and examined through test images based on three different convolutional neural networks, and classification results and success rate are reported.