A hybrid model of CNN and Transformer for medical image segmentation
Since medical images have the characteristics of low contrast,complex target shapes,and blurred edges,the segmentation accuracy of existing models cannot meet the requirements of high-precision modeling and automated surgery.In response,a hybrid architecture segmentation model called ParaCNNFormer was proposed,combining the excellent local feature extraction capabilities of convolutional neural networks(CNN)and the advantages of Transformer's long-distance modeling.As a U-shaped structure segmentation model,both the encoder and decoder of ParaCNNFormer adopted the hybrid architecture of CNN and Swin Transformer in parallel,which effectively improved the segmentation accuracy.CNN was used to extract local detailed features,and Swin Transformer was used to establish long-distance dependencies.The comparative experimental results on CHAOS and DSB18 datasets show that,the dice coefficient has been significantly improved compared with the popular TransUnet and SwinUnet.
medical image segmentationTransformerconvolutional neural networks(CNN)hybrid architecture
万方数据
维普
