Research on Medical Image Segmentation Method Based on Improved U-Net 3+

Liver disease is one of the most common diseases in medicine, and accurate segmentation is a necessary step to assist in the diagnosis and surgical planning of liver diseases. However, due to the complexity of liver CT images, liver segmentation remains a highly challenging problem. A segmentation model BAP-UNet3+based on the UNet3+network model is proposed to address the issue of low accuracy of existing convolutional neural networks in liver image segmentation. Introduce the biomimetic attention mechanism and point sampling method for target feature perception into the skip connections of UNet3+networks, and use a geometric contour loss function to perceive boundary information. The experimental results on the CHAOS dataset show that the average Dice reaches 0.9467, the average IoU reaches 0.962 3, and the average F1 Score reaches 0.9351, proving that the model can learn both image detail features and global structural features simultaneously, and can better segment liver images.