Multiple sclerosis lesion filling using a non-lesion attention based convolutional network

Multiple sclerosis (MS) is an inflammatory demyelinating disease of the central nervous system (CNS) that results in focal injury to the grey and white matter. The presence of white matter lesions biases morphometric analyses such as registration, individual longitudinal measurements and tissue segmentation for brain volume measurements. Lesion-inpainting with intensities derived from surrounding healthy tissue represents one approach to alleviate such problems. However, existing methods fill lesions based on texture information derived from local surrounding tissue, often leading to inconsistent inpainting and the generation of artifacts such as intensity discrepancy and blurriness. Based on these observations, we propose a non-lesion attention network (NLAN) that integrates an elaborately designed network with non-lesion attention modules and a designed loss function. The non-lesion attention module is exploited to capture long range dependencies between the lesion area and remaining normal-appearing brain regions, and also eliminates the impact of other lesions on local lesion filling. Meanwhile, the designed loss function ensures that high-quality output can be generated. As a result, this method generates inpainted regions that appear more realistic; more importantly, quantitative morphometric analyses incorporating our NLAN demonstrate superiority of this technique of existing state-of-the-art lesion filling methods.