Reliability-based robust multi-atlas label fusion for brain MRI segmentation
作者：
Highlights：
• We define two kinds of reliabilities for each voxel in the target image, including label reliability and spatial reliability, which are estimated based on the soft label and the spatial structure from the initial segmentation, respectively.
• We use the voxels with high reliabilities to help refine the label fusion process of those voxels with low reliabilities in the target image.
• We validate our proposed framework on four well-known label fusion methods (i.e., LWV, PBM, JLF and SPBM) and perform experiments on three public datasets (i.e., NIREP, LONILPBA40, and ADNI), with experimental results demonstrating the outperformance of our methods over three state-of-the-art methods in multi-atlas image segmentation.
摘要
•We define two kinds of reliabilities for each voxel in the target image, including label reliability and spatial reliability, which are estimated based on the soft label and the spatial structure from the initial segmentation, respectively.•We use the voxels with high reliabilities to help refine the label fusion process of those voxels with low reliabilities in the target image.•We validate our proposed framework on four well-known label fusion methods (i.e., LWV, PBM, JLF and SPBM) and perform experiments on three public datasets (i.e., NIREP, LONILPBA40, and ADNI), with experimental results demonstrating the outperformance of our methods over three state-of-the-art methods in multi-atlas image segmentation.
论文关键词：Label fusion,Label reliability,Spatial reliability,Multi-atlas segmentation,Brain structural MRI
论文评审过程：Received 4 December 2017, Revised 4 March 2019, Accepted 5 March 2019, Available online 8 March 2019, Version of Record 18 March 2019.
论文官网地址：https://doi.org/10.1016/j.artmed.2019.03.004
