| 摘 要: 使用机器学习方法对心磁数据样本有无疾病进行诊断分类。首先从心磁数据中提取除极阶段(TT间隔)的数据构建磁场图，然后求解电流密度图，从电流密度图中提取相关的磁场特征。针对非平衡数据分类问题，分别使用样本加权的SVM、LR、KNN、Adaboost和XGBoost五种学习模型进行训练，在此基础上设计了使用加权的LR和KNN为初级学习器、SVM为次级学习器的结合学习模型对样本数据进行训练。采用结合学习模型对73 名非患者和47 名心肌梗死患者的36 通道心磁数据进行实验，结果显示该模型对样本不均衡的心磁数据分类有较好的效果。 |
| 关键词: 心磁数据；电流密度图；机器学习；心肌梗死；结合学习 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金(61601173). |
|
| Magnetocardiograph Signal Diagnosis of Patients with Myocardial Infarction based on Machine Learning |
|
ZHAO Yongpeng, ZHU Junjie
|
(School of Electrical Engineering and Automation, Henan Polytechnic University, Jiaozuo 454000, China)
2959415512@qq.com; junjiezhu@hpu.edu.cn
|
| Abstract: This paper proposes to classify the presence or absence of diseases in magnetocardiograph data samples using machine learning method. First, magnetic field map, constructed by depolarization phase (TT interval) data, is extracted from the cardiac magnetic data. Then current density map is solved. From the current density map, the relevant magnetic field characteristics are extracted. Aiming at the problem of unbalanced data classification, five learning models of sample-weighted SVM (Support Vector Machine), LR (Logistic Regression), KNN (K-Nearest Neighbors), Adaboost and XGBoost are used for training. On this basis, weighted LR and KNN are designed as the primary learners, and weighted SVM as the secondary. The stacking model of the learners trains the sample data. The stacking learning model is used to conduct experiments on the 36-channel magnetocardiograph data of 73 non-patients and 47 myocardial infarction patients. The results show that the model has a good effect on the classification of unbalanced samples of the cardiogram data. |
| Keywords: magnetocardiograph data; current density map; machine learning; myocardial infarction; stacking learning |
