Automated Detection of Multiple Sclerosis Lesions in MRI with SegNet

Using deep neural networks, the proposed approach segments MRI images into classes and achieves good visual quality as well as similarity to the ground truth image. For the proposed process the dice similarity index is near 0.8. Dice similarity index is calculated for each class i.e. for WM, GM, CSF. Also the ground truth images are compared with the reference images and segmented MRI images that are used in the paper.