基于一类支持向量机的冠脉病变检测方法

发布时间：2018-04-03 07:36

本文选题：测量　切入点：冠脉病变检测　出处：《中国激光》2017年05期

【摘要】：针对冠脉病变检测算法普遍存在的异常截面识别率低、无法排除特殊结构影响等问题,提出了一种基于一类支持向量机(OCSVM)的冠脉病变检测方法,并使用冠脉面重采样和基于最大互信息的特征选择方法提高了算法识别正确率。该方法首先基于梯度通量对冠脉源截面进行三次样条插值重采样,然后构造出截面的多尺度特征,接着使用最大互信息结合冗余度去除进行特征选择,最后使用特征数据训练OCSVM完成冠脉病变检测。实验结果显示,在1128个冠脉截面数据的测试结果中,本算法在完全识别异常截面的情况下对健康截面的识别正确率达到了53.5%,远高于同类型的仅从正面和未标记数据学习的支持向量机(SVM)算法所对应的19.6%;而冠脉截面重采样也使得30个特征数下算法对健康截面的识别正确率由21.7%提高到了53.2%。
[Abstract]:In order to solve the problems of low recognition rate of abnormal cross section in coronary artery disease detection algorithm, which can not eliminate the influence of special structure, a new method of coronary artery lesion detection based on a class of support vector machine (SVM) is proposed.The recognition accuracy is improved by using coronary surface resampling and feature selection based on maximum mutual information.In this method, the coronary artery cross-section is firstly resampled by cubic spline interpolation based on gradient flux, then the multi-scale feature of the cross-section is constructed, and then the feature selection is carried out by using the maximum mutual information and redundancy removal.Finally, the characteristic data was used to train OCSVM to detect coronary artery lesions.The results of the experiment showed that, out of 1128 coronary artery cross-sectional data,The recognition accuracy of the healthy cross section is 53.5, which is much higher than that of the support vector machine (SVM) algorithm, which only learns from the positive and unlabeled data, and the coronary cross section is heavy.Sampling also increases the recognition accuracy of the health section from 21.7% to 53.2% under 30 feature numbers.
【作者单位】：天津大学精密仪器与光电子工程学院光电信息技术教育部重点实验室;中国人民解放军第二五四医院心血管内科;中国人民解放军第二五四医院放射科;
【分类号】：R541.4;TP18

【相似文献】