Graph-based semi-supervised one class support vector machine for detecting abnormal lung sounds
Graph-based semi-supervised one class support vector machine for detecting abnormal lung sounds
复制标题
DOI：
10.1016/j.amc.2019.06.001
复制
发表时间：
2020-01-01
影响因子：
4
通讯作者：
Liu, Guodong
中科院分区：
文献类型：
--
作者：
Lang, Rongling;Lu, Ruibo;Liu, Guodong
电子听诊器采集的异常肺音的检测在初级保健肺部疾病诊断和远程医疗中一般患者监测中起着基础作用。在过去的40年里，检测主要是通过监督学习进行的。然而，这种方法既费时又费钱，而且容易出错，因为它需要手工标记大量的样品。本文提出了一种基于图的半监督一类支持向量机（OCSVM）方法。它只需要使用少量标记的正常样本和大量未标记的样本作为训练样本，就可以描述正常的肺音，检测异常的肺音，避免了传统方法的缺点。构造了一个谱图来表示所有样本之间的关系，丰富了少量标记的正态样本所提供的信息。然后，建立了基于图的半监督OCSVM模型，并给出了求解方法。利用谱图中的信息，提高了识别和泛化的效果，这是有效检测异常肺音的关键。最后，通过在河北省石家庄市采集的所有样品进行实验，对所提出的方法进行了验证。实验结果表明，当标记样本较少时，该方法优于原始OCSVM。同时，该方法的性能随着未标记异常样本的增加而提高。(C) 2019由爱思唯尔公司出版。
The detection of abnormal lung sounds collected by electronic stethoscopes plays a fundamental role in pulmonary disease diagnostics for primary care and general patient monitoring in telemedicine. Over the past 40 years, the detection has been performed mainly by supervised learning. This method, however, is time- and cost-consuming, and error-prone because it requires manual labeling large numbers of samples. This work proposes a new method, a graph-based semi-supervised one class support vector machine (OCSVM). It can describe normal lung sounds and detect the abnormal ones only by using a small amount of labeled normal samples and abundant unlabeled samples as training samples, which avoids the shortcomings of the traditional methods. A spectral graph is constructed to indicate the relationship of all the samples, which enriches the information provided by only a small number of labeled normal samples. Then, a graph-based semi-supervised OCSVM model is built and its solution is provided. Employing the information in the spectral graph, the proposed method can enhance the effect of recognition and generalization which are crucial for the effective detection of abnormal lung sounds. Finally, the proposed method is evaluated by experiments with all the samples collected in Shijiazhuang, Hebei province, China. The experimental results show that the method outperforms the original OCSVM when the labeled samples are rare. Meanwhile, the performance of the proposed method becomes better as unlabeled abnormal samples increase. (C) 2019 Published by Elsevier Inc.