基于经食道超声心动图超声数据的心脏病分类

Classification of heart disease based on ultrasound data of transesophageal echocardiography

为了研究经食道超声心动图(TEE)超声数据与心脏病类别之间的关系,提出一种以决策树(DT)分类器作为基分类器的自适应提升(AdaBoost)分类预测模型(DT_AdaBoost)。该模型首先对训练集中的每个样本赋予一个相同的权重表示样本被选中的概率,然后有放回地选取样本组成训练子集训练DT分类器,如果该分类器的分类准确率大于50%,则计算此DT分类器的重要性,并更新样本权重,最后在新的样本分布下再次进行抽样训练。依此类推,可得到多个权重不同的DT分类器,把所有DT分类器按重要性叠加(boost)起来,即可得到最终的强分类器。仿真结果表明,以DT分类器作为基分类器的Ada Boost方法诊断准确率相对稳定在96.88%,高于以支持向量机(SVM)作为基分类器的94.70%、以K最近邻(KNN)作为基分类器的94.65%以及以朴素贝叶斯(Naive Bayes)作为基分类器的96.04%,并且较单一算法的分类器性能提高。

In order to study the relationship between Trans Esophageal Echocardiography（ TEE） ultrasound data and heart disease categories, an adaptive boosting（ AdaBoost） classification model（ DT_ AdaBoost） based on Decision Tree（ DT）algorithm was presented. Firstly, each sample in the training set was assigned with a same weight, which was on behalf of the probability that the sample was selected to the training subset. Secondly, a DT classifier was trained on the subset sampled from the training set. If the classification accuracy of this DT classifier was greater than 50%, then the importance of the DT classifier was calculated, and the sample weights were adjusted. Finally, another training subset was gained by sampling under the new sample distribution, and then another DT classifier was trained. By analogy, a series of DT classifiers that each of them had the different weight were built. The final strong classifier was got by superposing（ boost） all the DT classifiers according to their importance. The simulation results show that the classification accuracy of Ada Boost method using DT classifier as the basic classifier is relatively stable at 96. 88%, which is better than the 94. 70% of the Ada Boost method using Support Vector Machine（ SVM） classifier as the basic classifier, the 94. 65% of the Ada Boost method using K-Nearest Neighbor（ KNN） classifier as the basic classifier and the 96. 04% of the Ada Boost method using Naive Bayes classifier as the basic classifier. Besides, the performance of DT_ Ada Boost classifier is also improved than that of the classifier built by a single algorithm.