基于级联可分离空洞残差U-Net的肝脏肿瘤分割

计算机辅助肝脏肿瘤分割可减少医生工作量，提高手术成功率，因而具有重要的临床诊疗价值。为获得精确的肝脏肿瘤自动分割结果，该文结合医学影像分割领域近年新兴的U-Net模块提出了基于级联可分离空洞残差U-Net（cascaded separable and dilated residualU-Net，CSDResU-Net）的肝脏肿瘤分割方法。CSDResU-Net采用了级联操作，解决了因肿瘤在整幅图像中占比小而造成的肿瘤分割数据不平衡问题；通过在分割网络中整合残差单元、深度可分离卷积和空洞卷积，能够增加卷积核感受野并快速提取更具判别性的肝脏肿瘤图像特征，从而提高肝脏肿瘤分割精度。在国际医学图像计算和计算机辅助干预协会肝脏肿瘤分割数据库上的实验结果表明，CSDResU-Net比基线方法的Dice系数指标提升了1.3%，同时发现空洞率对分割网络的性能表现影响较大。

Cascaded Separable and Dilated Residual U-Net for Liver Tumor Segmentation

Computer-aided liver tumor segmentation can effectively reduce the workload of doctors and improve the success rate of surgery, and it has important clinical diagnosis and treatment value. Meanwhile, recently proposed U-Net model has achieved great success in the field of medical image segmentation. To obtain more accurate liver tumor segmentation results, this paper proposed an improved U-net model, i.e., cascaded separable and dilated residual U-Net (CSDResU-Net), for this medical application. CSDResU-Net utilizes cascade operation to solve the problem of unbalanced data in tumor segmentation due to the small proportion of tumors in the whole image. Besides, residual unit, depthwise separable convolution and dilated convolution are integrated into a single network to increase the convolution kernel receptive field, which can quickly extract more discriminative liver image features and lead to the performance improvement of liver tumor segmentation. Experimental results on the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) liver tumor segmentation (LiTS) benchmark dataset show that CSDResU-Net is relative to the baseline. The method improves the performance of the Dice coefficient by 1.3%, and at the same time proves that different void ratios have a greater impact on the performance of the segmentation network.