Automated Multiple Sclerosis Lesion Segmentation on MR Images via Mask R-CNN

Abstract

Multiple Sclerosis (MS) is a neurological disease with a remarkable incidence in young and middle-aged adults. When diagnosing MS on MR images, physicians often use computer-aided and automated secondary assistive tools in the decision-making process. Since the identification of MS lesions on MR images is a difficult and time-consuming process, performing MS lesions manually by experts can be prone to user error, variable and time consuming. In this study, a Mask R-CNN based deep learning method is proposed for automatic segmentation of MS lesions from MR scans. The MR image series used in the study are obtained from ISBI 2015 and MICCAI 2008 databases, which are publicly-available datasets. In the study, Detectron 2 framework is used as the infrastructure platform for architecture of Mask R-CNN. In experimental studies for automatic segmentation of MS lesions, Dice similarity scores of 86.30% and 81.32% are achieved on ISBI 2015 and MICCAI 2008 datasets, respectively. In conclusion, the Detectron 2-based Mask R-CNN deep learning method proposed in this study for automatic segmentation of MS lesions on MR slices is verified to be successful. © 2021 IEEE.