【摘要】 针对肝脏肿瘤存在的分割难点结合级联网络的思想,提出了一种融合了双注意力机制和U-Net架构优势的肝脏肿瘤分割网络(CDA-Net).首先,训练第一个DA-Net来实现肝脏的粗略分割;然后将第一阶段的分割结果与原始CT图做与操作,提取感兴趣区域,并将其输入第2个DA-Net实现肝肿瘤的精确分割;最后利用条件随机场对分割结果进行边缘约束,优化分割结果.在LiTS2017数据集上对模型进行训练及测试,平均Dice指标为0.658.实验结果表明,本文提出的方法具有较好的精度,证实了其对肿瘤分割的有效性.更多还原

【Abstract】 Aiming at the difficulty of liver tumor segmentation combined with the idea of cascade network a liver tumor segmentation network(CDA-Vet) is proposed that combines the advantages of dual attention mechanism and U-Vet architecture.First,the first DANet was trained to achieve rough segmentation of the liver.Then,the segmentation results of the first stage were combined with the original CT images to extract the region of interest and input it into the second DA-Vet to achieve accurate segmentation of the tumor.Finally,the edge constraint is applied to the segmentation results with conditional random field to optimize the segmentation results.The model was trained and tested on LiTS2017 data set,and the average Dice index was 0.6443.Experimental results show that the method proposed in this paper has good accuracy,which confirms its effectiveness for tumor segmentation.更多还原