论文部分内容阅读
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%。
Aiming at the low prevalence of abnormal cross-sectional recognition of coronary lesion detection algorithms and the inability to rule out the effects of special structures, a new type of coronary artery lesion detection method based on a support vector machine (OCSVM) And the feature selection method based on maximum mutual information improves the algorithm recognition accuracy. In the method, the cubic spline interpolation is resampled based on the gradient flux and then the multiscale features of the cross section are constructed. Then the maximum mutual information and redundancy removal are used to select the features. Finally, the feature data is used to train the OCSVM Coronary disease detection. The experimental results show that in the test results of 1128 coronary cross-sectional data, the correct recognition rate of the healthy cross-section is 53.5% when the abnormal cross-section is fully recognized, which is much higher than that of the same type of positive and unlabeled And 19.6% of support vector machine (SVM) algorithms for data learning. Coronary section resampling also improved the recognition accuracy of healthy sections from 21.7% to 53.2% with 30 features.