基于级联式三维卷积神经网络的肝肿瘤自动分割

目的:根据肝肿瘤CT影像中的特异性、分割难点以及残差网络思想，提出一种基于级联式卷积神经网络的全自动CT图像肝脏肿瘤分割方法。方法:首先根据临床知识对CT数据进行预处理，减少干扰；然后基于一个肝脏粗分割网络对肝脏进行分割，并根据分割结果坐标选取肝脏作为感兴趣区域；最后在感兴趣区域内对肿瘤进行精准分割。结果:通过级联式网络分割可以有效减少计算时间以及避免其它组织的干扰，从而实现肝肿瘤的快速分割。本研究提出的方法在2017年MICCAI肝肿瘤分割公开比赛数据集LiTS中进行测试，平均Dice分数为0.663，证实了其对肝肿瘤分割的有效性。结论:基于级联式卷积神经网络的全自动CT图像肝脏肿瘤分割方法可以实现肿瘤的快速分割。后期研究将继续增加数据量，对肿瘤进行分类，从而进一步完善模型。

Automatic liver tumor segmentation based on cascaded 3D convolutional neural network

Objective To propose a cascade convolutional neural network for full-automatic liver tumor segmentation in CT image according to the specificity of liver tumor CT image, difficulty of liver tumor segmentation and the idea of residual network. Methods Firstly, CT data were preprocessed based on clinical information, thus reducing interferences. Then a coarse liver segmentation network was used for liver segmentation, and according to the location of segmentation results, the liver was selected as the region of interest. Finally, the tumor was segmented accurately in the region of interest. Results A fast segmentation of liver tumor was realized by cascaded network segmentation which effectively reduced computational time and avoided interferences from other tissues. The proposed method was tested on the dataset of MICCAI 2017 liver tumor segmentation challenge (LiTS) and achieved an average Dice score of 0.663, which verified its effectiveness in the segmentation of liver tumor. Conclusion The full-automatic liver tumor segmentation in CT image based on cascaded convolutional neural network can be used to realize fast tumor segmentation. More cases will be included and tumor classification will be conducted in later studies, so as to further improve the model.