摘要
针对骨骼图像手动或半自动分割效率低的问题,提出了一种基于PE-Vnet网络的三维骨骼图像自动分割方法。该方法以骨骼CT图像序列作为输入,利用三维卷积提取骨骼的三维特征,结合3D Project&Excite(PE)模型对三维特征进行重标定,抑制了与骨骼分割任务无关的特征。设计了具有动态权重的损失函数,解决了训练样本数据前景和背景信息分布不均衡的问题。分别在自建数据集和Peréz-Carrasco等公开的数据集上进行了实验,分割结果的相似系数Dice值在自建数据集的股骨和胫骨实验分别为96.22%和95.68%,在公开数据集实验的Dice值为94.16%。实验结果表明,提出的方法在两个数据集上分割的结果皆接近于临床专家手动分割的结果,研究成果为骨科疾病的诊断和手术计划制定提供了新的参考。
Aiming at the problem that the low efficiency of manual or semi-automatic segmentation of bone images, a method for automatic segmentation of three-dimensional bone images based on PE-Vnet network is proposed. This method takes the bone CT image sequence as input, uses three-dimensional convolution to extract the three-dimensional features of the bone, and combines the 3 D Project & Excite(PE) model to recalibrate the three-dimensional features, which prevents the features that are not related to the bone segmentation task. A loss function with dynamic weights is designed to solve the problem of uneven distribution of foreground and background information in the training sample data. The experiments were conducted on the self-built data set and the data set published by Peréz-Carrasco et al. The similarity coefficient Dice value of the segmentation results is 96.22% and 95.68% in the femoral and tibial experiments of the self-built data set, and the Dice value in the public data set experiment is 94.16%. The experimental results show that the segmentation results of the proposed method on both data sets are close to the results of manual segmentation by clinical experts. The research results provide a new reference for the diagnosis of orthopedic diseases and the formulation of surgical plans.
作者
赵其杰
周安稳
朱俊豪
沈礼权
邵辉
Zhao Qijie;Zhou Anwen;Zhu Junhao;Shen Liquan;Shao Hui(School of Mechatronics Engineering and Automation,Shanghai University,Shanghai 200444,China;School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China;Microinvasive(Shanghai)Medical Robot Co.,Ltd.,Shanghai 200240,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2020年第7期243-251,共9页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61931022)项目资助
关键词
机器视觉
PE-Vnet网络
三维骨骼图像分割
全卷积神经网络
三维卷积
machine vision
PE-Vnet network
three-dimensional bone image segmentation
fully convolution neural network
three-dimensional convolution
