Dual-View Brain Tumor MRI Segmentation with Embedding Coordinate Attention Mechanism

In order to solve the problem of class imbalance affecting the segmentation accuracy due to the small proportion of brain tumor area in MRI, a dual-view MRI segmentation method embedded with coordinate attention mechanism is adopted. In the U-Net encoding path, the convolutional layer replaces the maximum pooling layer to reduce feature loss; Residual modules are added to the encoding and decoding paths to keep the transfer features intact; the decoding path is embedded with a coordinate attention mechanism, taking into account position information and channel relationships to alleviate class imbalance. The MRI is sliced along the axial and coronal views respectively, and the dual-view model is generated by fusion. Using spatial information. The research results show that the dice, sensitivity and precision evaluation indicators are significantly improved in the whole tumor area (WT), the core tumor area (TC) and the enhanced tumor area (ET). After fusion to generate dual views, the three areas Compared with the optimal method using the same data set, the dice index shows that WT is basically the same, and TC and ET are increased by 1.0% and 4.9% respectively. The research conclusion can effectively improve the MRI segmentation accuracy and has high reference and application value.