区域生长全卷积神经网络交互分割肝脏CT图像  

Region-growing fully convolutional neural network interactive segmentation of liver CT images

作　　者：张丽娟[1] 章润 李东明[2] 李阳[1] 王晓坤[3] ZHANG Li-juan;ZHANG Run;LI Dong-ming;LI Yang;WANG Xiao-kun(College of Computer Science and Engineering, Changchun University of Technology, Changchun 130012, China;School of Information Technology, Jilin Agricultural University, Changchun 130118, China;Aviation Operations Service College, Air Force Aviation University, Changchun 130022, China)

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130012 [2]吉林农业大学信息技术学院,吉林长春130118 [3]空军航空大学航空作战勤务学院,吉林长春130022

出　　处：《液晶与显示》2021年第9期1294-1304,共11页Chinese Journal of Liquid Crystals and Displays

基　　金：国家自然基金(No.61806024);吉林省教育厅科学研究项目(No.JJKH20181041KJ,No.JJKH20200678KJ,No.JJKH20210747KJ)。

摘　　要：由于医疗图像质量差、对比度低、患者之间差异大导致全自动分割方法很难获得足够准确、鲁棒的结果。为了解决全自动分割方法的局限性,本文提出一种基于神经网络改进的区域生长法,并与全卷积神经网络相结合对肝脏CT图像进行交互式分割。首先对图像进行预处理,突出待分割肝脏区域;接着计算像素在不同边缘检测算子下的梯度值作为该像素的特征,形成像素特征向量训练网络该网络以一对像素特征向量为输入,以两像素的关联度系数为输出;然后将训练好的神经网络模型作为区域生长算法的生长准则,手动交互选取一点产生分割结果;最后将分割结果作为原图的交互信息和原图灰度通道连接在一起一同输入全卷积神经网络。实验结果表明平均Dice系数达到96.69%,像素准确率达到99.62%,平均交并比达到96.65%。不同的腹部CT图像序列中肝脏的分割结果表明,该方法能精确提取肝脏区域,满足临床应用的需求。Due to poor medical image quality,low contrast,large differences between patients,it is difficult for fully automatic segmentation methods to obtain sufficiently accurate and robust results.In order to solve the limitations of the automatic segmentation method,this paper proposes an improved region growing method based on neural network,and combined with the fully convolutional neural network to interactively segment liver CT images.Firstly,the image is preprocessed to highlight the liver area to be segmented.Then,the gradient value of the pixel under different edge detection operators is calculated as the feature of the pixel to form a pixel feature vector training network.The network takes a pair of pixel feature vectors as input and the correlation coefficient of two pixels as output.Then,the trained neural network model is used as the growth criterion of the region growing algorithm,and a point is manually selected interactively to generate the segmentation result.Finally,the segmentation result is connected as the interactive information of the original image and the gray channel of the original image and input into the full convolutional neural network.The experimental results show that the average Dice coefficient reaches 96.69%,the pixel accuracy rate reaches 99.62%,and the average intersection ratio reaches 96.65%.The results of liver segmentation in different abdominal CT image sequences show that this method can accurately extract liver regions and meet the needs of clinical applications.

关 键 词：全卷积神经网络 区域生长法 交互式分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]