Research On Magnetocardiogram Signal Based On Machine Learning

Magnetocardiography(MCG)is a non-contact measurement of cardiac magnetic field signal on human body surface by using multi-channel superconducting quantum interferometer.It provides a new way for noninvasive study of cardiac electrophysiological activity by using MCG,which is of great significance for early diagnosis and screening of ischemic heart disease.Although magnetocardiogram has been used in clinical medicine for a long time,It is important for early inspection and viewing of ischemic heart disease.for instance,how to extract the abnormal characteristics of cardiac electrophysiology from magnetocardiogram of patients with heart disease,how to get better the method for noninvasive inspections of ischemic heart disease degree of sensitivity and specificity,and how to noninvasively locate the location of ischemic heart disease.This thesis on these three issues were discussed research.The following contents are included:(1)In this thesis,a method of average maximum current(AMDC)for noninvasive diagnosis of ischemic heart disease(IHD)with MCG is proposed From the visualization results of AMCD vector,we found that the four feature parameters extracted from AMCD showed obvious differences between patients and non patients.In the section of illness inspection,thisthesis summarizes five kinds of machine learning models with sample weighting,and uses these five kinds of weighted models to categorize the measured cardiac magnetic signals of ischemic heart disease with uneven samples.The classification effects of the unweighted machine learning model and the weighted machine learning model on the measured cardiac magnetic signals of ischemic heart disease with uneven samples are compared respectively.According to the results,I can get the conclusion We designed a intergrated learning model and studied the measured data.It turned out the weighted models is better than the model without weighting;the prediction effect of integrate learning model with sample weighting is better than that of single machine learning with sample weighting and single machine learning without sample weighting.(2)In the localization of ischemic heart disease,on the basis of the fuzzy c-means clustering(FCM)method,the sample size and the information entropy of features are integrated into the FCM objective function to obtain a new clustering objective function,thus an imporve FCM is proposed.Compared with FCM,This improved FCM clustering model has better clustering effect.According to the clustering results,the importance of the features is ranked,and the location of ischemic heart disease lesions is located by using the feature parameters sensitive to ischemic location.