Glioma segmentation algorithm based on hybrid offset axial self-attention mechanism
To improve the accuracy and quality of glioma MRI image segmentation,this paper designed a hybrid offset axial self-attention mechanism for the glioma segmentation algorithm ML A-Net(Multi-level axial-attention net).The offset axial self-attention mechanism and hybrid loss function designed in this algorithmic framework could be used to extract more accurate global relative po-sition relationships,as well as to enhance the sensitivity of the net to detailed structural features and to achieve the role of accurately segmenting the fuzzy boundaries of gliomas,respectively.The experimental results showed that the dice coefficient of MLA-Net could reach 0.843 3 and the Hausdorff distance was 2.587 on the mixed data of BraTS 2018 and 2019.The MRI image glioma seg-mentation performance of MLA-Net was excellent,and could fuse the global relative position features and local detail features to bet-ter segment the lesion region.
NETL
NSTL
万方数据
