Research On Assisted Diagnosis Of Coronary Artery Disease And Arrhythmia Based On Machine Learning

The morbidity and mortality of cardiovascular diseases in our country are increasing year by year,which seriously threatens people’s health.Early detection and early treatment can effectively reduce the morbidity and mortality of cardiovascular disease.How to achieve efficient,accurate,and low-cost detection of cardiovascular disease in the early stage of onset is a major challenge facing researchers.The ECG examination is the most commonly used non-invasive detection method in clinical practice.However,manual interpretation of the electrocardiogram not only needs to consider the subtle changes that occur in the ECG waveform,but also needs to consider complex and diverse clinical factors,which are time-consuming and laborintensive,and are not suitable for large-scale physical examination screening.Automatic ECG detection based on artificial intelligence methods provides a way to solve this problem.Existing research still has problems such as small samples,inconsistent data labeling,poor model interpretability and poor generalization ability.In order to solve the above problems,this paper focuses on ensemble learning,attention mechanism and comparative learning to carry out theoretical research,mainly research work and innovation are as follows:In view of the poor generalization ability of a single model in the existing assisted diagnosis of coronary heart disease,an ensemble learning method based on correlation coefficients is proposed.This method first selects a base-level model based on the correlation coefficient between models,and then uses the enumeration method to search for the optimal combination of the base-lMedical Devices.Woodhead Publishing,2019:891-911.evel model,and the output of the base-level model is used as the input feature of the meta-level model.Finally,the output results of the base-level model are integrated through the meta-level model to construct an ensemble learning model.In addition,the ensemble model is used to verify the reliability of the model on multiple public data sets.The results show that the ensemble learning model improves the accuracy of assisted diagnosis of coronary heart disease,achieving accuracy,sensitivity and specificity of 95.4%,95.8%and 94.4%,respectively.Compared with the existing research results,this method is better than a single classifier and has stronger generalization ability.Aiming at the deep learning model used in the existing arrhythmia-assisted diagnosis method without considering the correlation between the ECG signal amplitude,time interval and leads,a non-local convolutional block attention module is proposed.This module combines time,space and channel attention mechanisms to improve feature representation ability.A weight matrix is constructed in the model to represent the correlation between time,space,and channel attention mechanisms.Comparative analysis is carried out between the model and clinical diagnosis basis,which improves the credibility of the model.The proposed non-local convolutional block attention mechanism achieves an average accuracy of 98.6%for 5 types of assisted diagnosis of arrhythmia,and an AUC of 93.14%for multi-label assisted diagnosis of arrhythmia.The results show that the non-local convolutional block attention mechanisms have better performance than the model based on a single attention mechanism.The fusion of multiple attention mechanisms in assisted diagnosis of arrhythmia has a broader clinical application prospect.Aiming at the problem that different ECG databases cannot be assembled to train classification models due to inconsistent expert annotations,research is carried out based on comparative learning and a supervised learning model is proposed that intelligently recognizes ECG features,classifies arrhythmias based on ECG features,and checks and adjusts expert annotation databases.First,7 public databases are used and a more efficient feature representation model based on contrastive predictive coding is designed through ablation experiment.In addition,the data set is augmented with randomly generated noise,thereby increasing the difficulty of the training task.Then the pre-trained model is finetuned on multiple large data sets with labels in order to verify the validity of the model.Finally,the model is visualized to explain in depth the correspondence between the different prediction results output by the model and the clinical diagnosis.The improved comparative learning model in this paper has achieved an AUC of 94.39%in the multi-label assisted diagnosis of arrhythmia.Experimental results show that this method improves the model’s ability of ECG feature representation.It also improves the accuracy of the diagnosis of 12 types of arrhythmia.Compared with the existing research,the model eliminates the unnecessary network layer and is more streamlined and has stronger application prospects.