Localization of Demyelinating Plaques in MRI using Convolutional Neural Networks

Bartlomiej Stasiak, Pawel Tarasiuk, Izabela Michalska, Arkadiusz Tomczyk, Piotr S. Szczepaniak

Abstract: In the paper a method of demyelinating plaques localization in head MRI sequences is presented. For that purpose a convolutional neural network is used. It is trained to act as non-linear filter, which should indicate (give a high response) in those image areas where the sought objects are located. Consequently, the output of the proposed architecture is an image and not a single label as it is in the case of traditional networks with pooling and fully connected layers. Another interesting feature of the proposed solution is the ability to select network parameters using smaller patches cut from training images which reduces the amount of data that must be propagated through the network. It should be emphasized that the conducted research was possible only thanks to the manually outlined plaques provided by radiologist.