基于U型全卷积神经网络的腹部动脉CT图像分割算法研究  

Research on CT Image Segmentation Algorithm of Abdominal Artery Based on U-Shaped Fully Convolutional Neural Network

作　　者：赵秀兰 刘印文 ZHAO Xiulan;LIU Yinwen(CT Room,Haoji Center Health Center,Leping Town,Chiping County,Liaocheng Shandong 252126,China;CT Room,Hospital of Traditional Chinese Medicine of Dongchangfu District,Liaocheng Shandong 252003,China)

机构地区：[1]茌平县乐平镇郝集中心卫生院CT室,山东聊城252126 [2]聊城市东昌府区中医院CT室,山东聊城252003

出　　处：《中国医疗设备》2021年第2期85-88,共4页China Medical Devices

摘　　要：目的为了精确分割腹部动脉血管,提出一种基于深度学习的全自动腹部动脉CT图像分割算法。方法采用区域不平衡块生成方法提取CT血管横断面、冠状面和矢状面图像特征,接着采用U型全卷积神经网络对块特征进行训练与分割,最后采用最大体素保留法获得三维血管分割图像。选用120例患者腹部CT血管图像进行网络训练和分割实验,分割结果评价指标采用精确率、召回率和Dice系数。结果基于U型全卷积神经网络能分割全部腹部CT图像大血管和绝大多数小血管。全卷积神经网络中块尺寸s=32所得平均Dice系数、精确率和召回率分别达87.2%、85.9%和88.5%,且与块尺寸s=48和s=64大致相等。基于U型全卷积神经网络所得平均Dice系数、精确率和召回率均优于其他血管分割算法。结论基于U型全卷积神经网络算法的图像分割精度高,是一种可行的腹部CT血管分割算法。Objective To propose an automatic CT image segmentation algorithm of abdominal artery based on deep learning,in order to segment abdominal artery accurately.Methods The CT image features of cross section,coronal plane and sagittal plane of vessels were extracted by using the method of area imbalance patch generation.Then,the U-shaped fully convolutional neural network(U-Net)was adopted to train and segment the patch features.Finally,the three-dimension segmentation image was achieved by using the maximum voxel preservation method.120 cases of abdominal CT vascular images were selected for network training and segmentation experiments.The segmentation accuracy was measured by precision rate,recall rate,and Dice coefficient.Results All large vessels and most small vessels in abdominal CT images could be segmented based on U-Net method.The average Dice coefficient,precision rate,and recall rate of U-Net with patch size s=32 were 87.2%,85.9%and 88.5%respectively,which were approximately equal to patch size s=48 and s=64.Moreover,the average Dice coefficient,precision rate,and recall rate of the proposed method based on U-Net were better than other vessel segmentation algorithms.Conclusion The segmentation accuracy of image based on U-Net method is high,which is a feasible abdominal CT vascular segmentation algorithm.

关 键 词：图像分割 全卷积神经网络 U型网络 计算机断层显像 动脉 

分 类 号：R318[医药卫生—生物医学工程] R814.4[医药卫生—基础医学]