VCMix-Net: A hybrid network for medical image segmentation

With the continuous development of convolutional neural networks, their applications in medical image seg-mentation have become increasingly widespread. By using techniques such as channel attention and spatial attention, better segmentation results have been achieved. However, most existing networks lack the ability to select informative features when segmenting medical images. The combination of channel and spatial attention assumes that the interaction between channel and spatial information is independent, which can lead to biases and affect the model's performance. In certain cases, important information is assigned less weight, resulting in lower accuracy of small sample segmentation in large backgrounds due to the interdependence of this interac-tion. To address this issue, this paper proposes the Vip and Convolution Mixed (VCMix) module. It utilizes parallel operations of convolution and a multi-layer perceptron with class attention. By employing tensor shifting and linear projection, the module simultaneously captures local information and local-global information. It not only reduces the interdependence between channel and spatial information but also leverages the advantages of convolution in capturing local information to correct biases in local-global information, thereby obtaining more accurate feature information. The VCMix module can be integrated into the U-Net architecture. The model is evaluated on three datasets: LiTs, Lung, and ISIC-2016. Experimental results demonstrate the excellent perfor-mance of the VCMix module on all three datasets, highlighting its effectiveness in medical image segmentation. The parallel operations of the VCMix module provide insights for parallel operations of convolution with other methods, which contribute to the accurate delineation of lesion areas in medical image segmentation. Furthermore, it lays a foundation for the integration of artificial intelligence techniques from different domains and the development of AI in the medical field.