Research And Application Of Improved Model Based On UNet Model In Medical Image

At present,in-depth learning in the field of medical image segmentation has become more and more mature.Through the use of computer,the established depth learning model can fully help doctors to carry out medical image segmentation.Because the U-shaped structure designed by UNet has excellent performance in the field of medical image segmentation,most of the current depth learning models are based on UNet.The U-shaped structure and jump connection layer of UNET can effectively achieve accurate image segmentation.However,for complex images,the network structure of UNet alone is not sufficient.As the technology continues to update,more variants of the UNET model appear,they have achieved very good results.Among them,UNET + + is an excellent model in UNET variety.In the design of UNet + +,scholars have added a more intensive jump connection layer in U-Net.Compared with U-Net,UNet ++ is more efficient in processing complex images.In this paper,channel attention mechanism and spatial attention mechanism are used to improve the UNET + + model to obtain better efficiency and accuracy of image segmentation.At the same time,two new network models of medical image segmentation are designed based on UNet + +.A-UNet + + and CA-UNet + +,respectively.A-UNet + + uses a spatial attention module to address the loss of eigenvalues during up-sampling for lung medical image segmentation in the context of simple medical images.CA-UNet + + uses the channel attention module and the spatial attention module to solve the eigenvalue loss in the process of longdistance jumping connection and up-sampling respectively,aiming at the liver medical image segmentation under complex background.Experimental results and data analysis show that compared with Attention-UNet,RU-Net,UNet + +,the proposed ca-UNet + + and a-UNet + +can achieve better performance in medical CT image segmentation.