Research On Detection Of Myocardial Ischemia Based On Deterministic Learning And Multi-feature Fusion

With the economic development,cardiovascular disease,especially ischemic heart disease(referred to as myocardial ischemia)has become the main cause of death in our country,seriously threatening people’s life and health.Electrocardiogram(ECG)is a non-invasive,universal,convenient and low-cost tool for the initial diagnosis of heart disease.However,accurate diagnosis based on ECG requires doctors to have extensive clinical experience,and the diagnostic accuracy rate is not high(about 60%).Therefore,in-depth analysis of ECG using the latest technology of artificial intelligence is of great significance to improve the diagnostic accuracy of myocardial ischemia.Cardiodynamicsgram(CDG)is a new method for detecting myocardial ischemia obtained by dynamic modeling of ECG signals using deterministic learning(a new machine learning method in the dynamic environment)in recent years.However,CDG is in the early stage of development and needs in-depth research.Therefore,based on the existing work,this paper further explores the diagnosis of myocardial ischemia by CDG,and achieved the expected results,as follows:(1)Based on the existing work,this paper proposed an improved CDG(Improved Cardiodynamicsgram,ICDG)for myocardial ischemia detection.Existing work only modeled the ST-T segment(ST segment and T wave)of the ECG signal to obtain CDG.However,clinical studies have shown that in addition to the ST-T segment reflecting cardiac repolarization,the QRS complexes reflecting cardiac depolarization also contain rich ischemia-related information.Therefore,this paper proposed an improved method,that is,using a deterministic learning algorithm to model the dynamics of the complete ECG cycle(repolarization and depolarization)data,and obtained the ICDG describing the whole process of cardiac changes;Further,this paper mined the features useful for myocardial ischemia detection from ICDG,and extracted 12 new spatiotemporal features;in addition,this paper extracted 220 classical ECG spatiotemporal features from 12-lead ECG,and fused ICDG features with classical ECG features,for better myocardial ischemia detection performance.(2)This paper collected data from different sources of myocardial ischemia cases and nonischemic individuals to establish a clinical dataset.A total of 333 ECGs were collected from two medical institutions;including 141 myocardial ischemia patients and 192 non-ischemic individuals.Different medical institutions use different models of electrocardiographs to acquire 12-lead ECGs,which results in incompatible ECG files.Therefore,this paper first considers a standard to standardize ECG files from multiple sources to provide data in a unified format for subsequent research.(3)The performance of the proposed ICDG features and multi-feature fusion in detecting myocardial ischemia was evaluated on the self-built clinical dataset.First,the ability of ICDG features to detect myocardial ischemia was assessed.The 5 most discriminative features were selected from 12 ICDG spatio-temporal features using a feature selection method.The support vector machine(S VM)classifier was trained with 5-fold cross-validation,and achieved average accuracy of 88.83%,sensitivity of 86.80%,specificity of 90.30%and areas under the receiver operating characteristic curve(AUC)of 0.93,which outperforms ECG features.Further,the effect of fusing ICDG feature and ECG feature on performance of myocardial ischemia detection was evaluated.The 16 most important features were selected from 220 ECG features using a feature selection method,and fused with the above 5 ICDG features to construct a vector of features,which achieved better performance than any single type of feature,with average accuracy of 91.38%,sensitivity of 92.73%,specificity of 90.44%and AUC of 0.96.The experimental results showed that the method based on deterministic learning and multi-feature fusion can make full use of the multi-level ischemia information of ECG,and obtain better performance in myocardial ischemia detection.The ICDG and multi-feature fusion method proposed in this paper is expected to provide a convenient and effective tool for the accurate diagnosis of myocardial ischemia,and has important clinical application value.