融合多注意力机制的脊椎图像分割方法.

Considering the existing spinal Computer Tomography（CT） and Magnetic Resonance（MR） image segmentation models have limitations in segmentation performance, this paper proposes a spinal segmentation method MAU-Net based on U-shaped network. Frist, this paper introduces coordinate attention module into the encoder of Ushaped network, which enables the network accurately capture the spatial position information and embed it into the channel attention. Second, this paper proposes dual-branch channel cross fusion module based on Transformer, it can replace the skip connection for multi-scale feature fusion. Finally, this paper proposes a feature fusion attention module to better fuse the semantic differences between Transformer and convolution network. On scoliosis CT dataset, Dice reached 0.9296, IoU reached 0.8597. On the public MR dataset SpineSagT2 Wdataset3, compared with FCN, Dice improved by 14.46%. Experimental results show that this method can effectively reduce the false segmentation area of vertebrae. [ABSTRACT FROM AUTHOR]