Coronary heart disease (CHD) is a typical cardiovascular disease whose occurrence and development is a long process. Timely and accurate diagnosis of patients with varying degrees of coronary artery stenosis (VDCAS) is conducive to accurate treatment and prognosis assessment. This study aims to correctly classify VDCAS patients by utilizing multi-domain features fusion of single-lead 5-min ECG signals and machine learning methods, so as to provide reference for doctors to judge the CHD development process. ECG signals were collected from 206 subjects with CHD, mild CHD, thoracalgia and normal coronary angiograms (TNCA), and healthy. Then, the time, frequency, time–frequency, and nonlinear domain features of ECG signals were extracted to establish a multi-domain feature set. To get the optimum subset of features, the recursive feature elimination (RFE) and information gain (IG) were selected. Subsequently, eXtreme Gradient Boosting (XGBoost) and random forest (RF) were adopted for classification. Results indicated that RFE combined with XGBoost was significantly effective in classifying VDCAS patients. When the four categories of subjects (CHD, mild CHD, TNCA, and healthy) were classified, the average accuracy, sensitivity, specificity, and F1-score of the proposed method were 91.74%, 89.39%, 96.80%, and 90.09%, respectively. Besides, three categories of subjects (no stenosis, luminal narrowing [Formula: see text] 50%, and luminal narrowing [Formula: see text] 50%) and two categories of subjects (CHD and healthy) were also analyzed, and the average accuracy was 91.27% and 98.46%, respectively. The results suggest that the proposed method can provide reference for doctors to judge VDCAS patients.

中文翻译：

---

一种基于心电图信号检测冠状动脉狭窄的方法

冠心病（coronary heart disease，CHD）是一种典型的心血管疾病，其发生和发展是一个漫长的过程。对不同程度冠状动脉狭窄（VDCAS）患者进行及时准确的诊断，有利于准确的治疗和预后评估。本研究旨在利用单导联5分钟心电信号的多域特征融合和机器学习方法对VDCAS患者进行正确分类，为医生判断冠心病发展过程提供参考。从 206 名患有冠心病、轻度冠心病、胸痛和冠状动脉造影（TNCA）正常和健康的受试者收集心电图信号。然后，提取心电信号的时域、频域、时频域和非线性域特征，建立多域特征集。为了得到最优的特征子集，选择递归特征消除（RFE）和信息增益（IG）。随后，采用eXtreme Gradient Boosting（XGBoost）和随机森林（RF）进行分类。结果表明，RFE 联合 XGBoost 对 VDCAS 患者的分类效果显着。当对四类受试者（CHD、轻度 CHD、TNCA 和健康）进行分类时，所提出方法的平均准确度、灵敏度、特异性和 F1-score 分别为 91.74%、89.39%、96.80% 和 90.09%，分别。此外，还分析了三类受试者（无狭窄、管腔狭窄 [公式：见文] 50%、管腔狭窄 [公式：见文] 50%）和两类受试者（冠心病和健康），平均准确率分别为 91.27% 和 98.46%。