基于深度学习的颈动脉超声图像内中膜厚度测量  

Carotid intima-media thickness measurement in ultrasound image based on deep learning

作　　者：孙萍[1] 李锵[1] 关欣[1] 滕建辅[1] 

机构地区：[1]天津大学电子信息工程学院,300072

出　　处：《国际生物医学工程杂志》2016年第5期257-262,I0003,I0004,共8页International Journal of Biomedical Engineering

基　　金：国家自然科学基金（61472163）

摘　　要：目的 颈动脉血管内中膜厚度（IMT）是衡量动脉粥样硬化程度的重要标准.一般采用人工标定进行测量,该过程耗时且繁琐,由此提出一种总体性能较好的全自动分割（AS）算法.方法 该算法首先利用卷积神经网络（CNN）识别出颈动脉血管远端,进而提取包含颈动脉内膜、中膜部分的感兴趣区域（ROI）.采用基于堆栈式自编码器（SAE）构造的模式分类器将ROI中的像素进行分类.最后利用分类区域的面积信息和位置信息对分类结果进行甄别,运用曲线拟合提取边界完成测量任务.结果 针对本研究所用图像库中的84幅颈动脉超声图像进行实验,金标准（GT）由两名专家4次测量的平均值产生,其与AS之间的绝对误差和标准差为（13.3±20.5） μm,协方差系数为0.990 7.结论 实验结果表明,此算法总体性能较好,能够实现超声颈动脉血管内中膜全自动、快速、准确分割,从而满足临床需要.Objective The common carotid artery intima-media thickness （IMT） is a widely accepted and important marker of early atherosclerosis,and measurement of IMT based on manual tracing is time-consuming and complicated.A fully automatic segmentation （AS） method was proposed in this study for the IMT measurement to overcome the drawbacks.Methods First,convolutional neural network （CNN） was applied to identify carotid artery distal and region of interest （ROI） was extracted,which included intima-media complex （IMC）.The pattern classifier based on the stacked auto encoder （SAE） was added to classify pixels of ROI.Reliable classification regions were chosen based on region area and region center,and the IMT measurement was completed by extracting final boundary with the method of curve fitting.A total of 84 ultrasound images from 84 corresponding patients were tested with the proposed method.The ground truth （GT） of IMT was manually measured for four times by two experts and then averaged,and the automatic segmented IMT was computed using the proposed method.Results The mean of the absolute error and standard deviation between AS and GT IMT was （13.3±20.5） μm,and the correlation coefficient was 0.990 7.Conclusions Experimental results show that the over all performance of the proposed method is better,and it can achieve automatic,fast and accurate segmentation of intima-media of common carotid artery,which satisfy the clinical requirements.

关 键 词：内中膜厚度 图像分割 深度学习 卷积神经网络 堆栈式自编码器 

分 类 号：R543.5[医药卫生—心血管疾病] TP391.41[医药卫生—内科学]