Early Detection And Evaluation Of Ischemic Heart Disease Based On Deterministic Learning

Ischemic heart disease is one of the most important causes of death worldwide.Early detection of ischemic heart disease is a key factor in realizing timely and effective treatment,reducing or even avoiding myocardial infarction caused by persistent myocardial ischemia,and saving lives.Electrocardiography(ECG)is currently the simplest and most widely used clinical method for diagnosing cardiovascular diseases.However,many patients with myocardial ischemia still have normal or roughly normal ECGs in clinical practice,and the diagnostic accuracy of myocardial ischemia is not high.Early and accurate detection of myocardial ischemia based on ECG is still an important and difficult problem in the field of cardiovascular disease.Cardiodynamicsgram(CDG)is a new method of electrocardiograph(ECG)analysis for myocardial ischemic that we have proposed in recent years.CDG is the threedimensional visualization of the subtle cardiac dynamics related to myocardial ischemia extracted from the ST-T segments of ECG using deterministic learning(a machine learning method in the dynamic environment).A clinical trial about myocardial ischemia detection was carried out using CDG in Fuwai Hospital Chinese Academy of Medical Sciences.For patients with suspected coronary heart disease whose ECG is roughly normal,ECGs are labeled on the basis of coronary angiography.The results showed that CDG could detect patients with coronary heart disease more accurately when ECG is normal or roughly normal.This study further adopted the sampling deterministic learning algorithm to carry out early detection and evaluation of ischemic heart disease from the perspective of system identification and nonlinear dynamics,mainly including the following work:1)An interpretable model for detecting myocardial ischemia was established based on deterministic learning and CDG.Coronary anthography was considered as the ”gold standard” for diagnosing coronary heart disease(or myocardial ischemia)in pre-trial at Fuwai Hospital.However,coronary angiography is only the ”gold standard” for evaluating coronary artery stenosis,cannot directly assess the severity of myocardial ischemia.In addition to coronary stenosis,slow coronary blood flow is also an important cause of myocardial ischemia.In response to this problem,firstly,we collected cases at Fuwai hospital and Shihezi people’s hospital and constructed a representative multi-medical center myocardial ischemic ECG dataset,including both cases with ischemic ECG changes,normal or roughly normal ECGs,as well as coronary artery occlusive lesion and nonocclusive lesion verified by coronary angiography.Secondly,the CDG of the patient with roughly normal ECG was calculated and the stenosis group and the non-stenosis group were randomly divided into the unrelated training set and testing set.Then,a myocardial ischemia model using the support vector machine was trained in the training set and the model accuracy was evaluated by using the test set.We further used interpretable CDG to analyze the false-positive cases in turn and found that most of them had slow blood flow phenomena(a non-obstructive coronary artery disease).The slow blood flow cases were relabelled to improve the accuracy of myocardial ischemia data labeling.Finally,on this basis,we constructed a more accurate and interpretable model of myocardial ischemia detection based on deterministic learning and CDG.2)A non-invasive and convenient tool for evaluating the effects of percutaneous coronary intervention(PCI)on acute coronary syndrome(ACS)was established based on deterministic learning.PCI is an effective strategy to recanalize the narrowed or occluded epicardial vessels sustainedly for revascularization.However,non-invasive and precise measurements for monitoring the efficacy of PCI remain scarce.In response to this problem,we further explored the value of CDG in evaluating the effect of revascularization in ACS patients undergoing PCI based on the research that CDG can effectively detect myocardial ischemia.First,ACS patients undergoing successful PCI were enrolled in this study.12-lead ECGs were recorded before PCI(pre-PCI)and after PCI(post-PCI).CDG was extracted by deterministic learning from ST-T segments in ECG.The QT parameters of ECG were recorded to compare with CDG.The results showed that after successful PCI,CDG decreased significantly,and its morphology gradually changed from scattered to regular,but the QT parameters did not change significantly.The results suggested that CDG reflects the dynamic changes of myocardial ischemia before and after PCI accurately and intuitively,and improves the ability of ECG signal sensitivity to identify blood perfusion,which may be a convenient and non-invasive auxiliary tool for clinical evaluation of the efficacy of revascularization.3)We further proposed more effective quantitative morphological features of ECG and evaluated the early detection ability of CDG for myocardial ischemia in multi-medical centers.Previous studies have shown that the myocardial ischemia detection model established by the Lyapunov index and the spectrum fitting index describing the shape of the CDG can achieve more accurate detection of myocardial ischemia.However,in some cases,the Lyapunov index cannot accurately characterize the irregularity of CDG.Therefore,we further use the complexity analysis method-Lempel-Ziv(LZ)complexity to improve the characterization of CDG.For ischemic patients with roughly normal ECGs in Beijing Fuwai Hospital and Shihezi People’s Hospital,it was found that that there were significant differences in Lyapunov index and LZ complexity between the myocardial ischemia group and the non-ischemic group,and the model based on LZ complexity and the spectral fitting index was more effective for the detection of myocardial ischemia.However,in some cases,it is not easy to evaluate which of the two features of the Lyapunov index and LZ complexity describes CDG more accurately.Therefore,we proposed a classification method based on ensemble learning to fuse different CDG features to construct a robust and generalized myocardial ischemia detection model.Finally,the generalization ability and robustness of the model were further verified on the dataset of Beijing Fuwai Hospital,Shihezi People’s Hospital,and Qilu Hospital of Shandong University,as well as on the public dataset PTB.The results showed that the ischemic detection model proposed in this manuscript provided interpretable classification features and a clinically consistent decision-making process,making it easy to understand and accept for clinicians.More importantly,the results in the clinical environment proved that the ischemic model based on CDG is with good generalizability on new unseen patients.In summary,this study finally constructed a representative multi-medical center ischemic heart disease ECG dataset that is close to the clinical environment,there are both cases with ischemic ECG changes,normal or generally normal ECG,as well as coronary artery occlusive lesion and non-occlusive lesion verified by coronary angiography,which provides valuable data resources for subsequent related researches.Furthermore,an interpretable ischemic detection model is constructed with accurately labeled multicenter samples,providing an effective and convenient tool to assist clinicians in diagnosing myocardial ischemia timely and evaluating the treatment effect of patients with ischemic heart disease quickly and effectively.