改进U-Net融合空洞卷积的肝脏计算机断层扫描影像分割算法

Liver Computed Tomography Image Segmentation Algorithm by Improved U-Net Fused with Dilated Convolution

为了提高肝脏计算机断层扫描(CT)影像分割的精度并解决边缘修正不平衡问题,提出一种基于改进U-Net并融合空洞卷积的肝脏CT影像分割算法。通过引入改进注意力特征机制模块增强全局信息,将传统的空洞卷积分解为一维卷积,并结合残差连接来强化上下文信息。使用编码器筛选U-Net中的图像信息,将改进的U-Net模块与空洞卷积模块融合,并通过混合池化层进行图像分割。在医学图像分割十项全能肝脏数据集上的实验结果表明,该算法在保留肝脏CT影像边缘信息的精度上优于其他模型,系数D和Q分别为93.98%和96.74%,平均分割时间仅57 ms。

To improve the accuracy of liver CT image segmentation and address the issue of unbalanced edge correction,a liver CT image segmentation algorithm based on an improved U-Net and fused dilated convolutions is proposed.This algorithm aims to resolve the mentioned problems.It employs an enhanced global information module with an improved attention feature mechanism,decomposes traditional atrous convolution into one-dimensional convolution,and integrates residual connections to strengthen contextual information.The encoder is used to filter image information in the U-Net network,and the improved U-Net module is fused with the atrous convolution module to achieve image segmentation through a mixed pooling layer.Experimental results on the MSD liver dataset show that the proposed algorithm outperforms other models in preserving the accuracy of edge information in liver CT images,with a D coefficient of 93.98%and a Q coefficient of 96.74%.The average segmentation time is only 57 ms.