2020/10/19 by Zicong Zhang, Zhang, Zicong, Kimerly Powell +11
Medicine · Computer Science · #Acute Ischemic Stroke Management #Medical Image Segmentation Techniques #Cerebrovascular and Carotid Artery Diseases
paper · pdf · doi:10.48550/arxiv.2010.09586
White Matter Hyperintensities (WMH) are the most common manifestation of\ncerebral small vessel disease (cSVD) on the brain MRI. Accurate WMH\nsegmentation algorithms are important to determine cSVD burden and its clinical\nconsequences. Most of existing WMH segmentation algorithms require both fluid\nattenuated inversion recovery (FLAIR) images and T1-weighted images as inputs.\nHowever, T1-weighted images are typically not part of standard clinicalscans\nwhich are acquired for patients with acute stroke. In this paper, we propose a\nnovel brain atlas guided attention U-Net (BAGAU-Net) that leverages only FLAIR\nimages with a spatially-registered white matter (WM) brain atlas to yield\ncompetitive WMH segmentation performance. Specifically, we designed a dual-path\nsegmentation model with two novel connecting mechanisms, namely multi-input\nattention module (MAM) and attention fusion module (AFM) to fuse the\ninformation from two paths for accurate results. Experiments on two publicly\navailable datasets show the effectiveness of the proposed BAGAU-Net. With only\nFLAIR images and WM brain atlas, BAGAU-Net outperforms the state-of-the-art\nmethod with T1-weighted images, paving the way for effective development of WMH\nsegmentation. Availability:https://github.com/Ericzhang1/BAGAU-Net\n