Medical Image Segmentation Network Integrated with CNN-Transformer
Mainstream convolutional neural networks usually face three challenges in medical image segmentation.First,con-ventional convolutional operations mainly acquire local features of medical images,which show limitations in image long-range in-formation modeling capabilities.Secondly,the conventional downsampling operation in convolutional neural networks will lead to the loss of important information in the feature map of medical images,affecting the segmentation effect.Finally,when the problems caused by convolutional operations are solved,how to fully integrate the extracted local features and global features.In order to solve the above problems,a medical image segmentation network integrating CNN and Transformer is proposed.The network first solves the problem of convolutional operation sensor field fixation by introducing Transformer.Second,Patch Embedding is used to solve the problem of important information loss during the downsampling process.Finally,the problem of insufficient fusion of local and global features is solved by alternating cnns and transformers.Experimental results on the ISIC2018 and KiTS19 datasets show that the proposed network is not only able to capture finer contour arcs,but also has strong anti-interference ability,with high segmenta-tion accuracy and robustness.
deep learningconvolutional neural networkssemantic segmentation of medical imagesTransformer
万方数据
维普
