Study And Construction Of A System To Assist In The Diagnosis Of Coronary Heart Disease Based On Multimodal Medical Data

Coronary heart disease is a common cardiovascular disease,which is caused by the narrowing or blockage of coronary arteries leading to myocardial ischemia and damage,and is one of the main causes of cardiovascular disease death.Electronic medical records and heart sound signals contain a large amount of biomedical data related to heart health,so analyzing these two signals is expected to provide possibilities for early non-invasive detection of coronary heart disease.Most machine learning-based studies focus on the analysis of a single modality signal,but baseline data and heart sound signals are interrelated,and their complementary relationship can provide more comprehensive and accurate information to evaluate patients’ heart conditions.In addition,existing studies mainly use traditional features or deep learning features for analysis,and there are few studies that combine multiple types of features for diagnosis.This article uses clinically synchronized coronary heart disease electronic medical records and heart sound data to explore the application value of using multimodal signals and multiple types of features for early non-invasive detection of coronary heart disease.In this study,we diagnosed coronary heart disease based on single modality baseline data and heart sound signals respectively,and proposed an integrated deep learning method based on the joint analysis of multimodal signals,aiming to explore the application value of using multimodal signals and multiple types of features for early non-invasive detection of coronary heart disease.The main work and innovation points of this article are as follows.(1)Regarding heart sounds,this article uses a heart sound denoising method based on a 6th order Butterworth filter,followed by synchronous wavelet compression processing based on the time-frequency characteristics of each component of the signal,and combined with the SE-Res Ne Xt-50 network structure,the time-frequency graph of the synchronous wavelet transform is used as the input of the model.Experimental results show that this method has a high accuracy,precision,and recall in coronary heart disease detection,reaching 84.81%,84.38%,and 84.25%,respectively.(2)Regarding structured baseline data,this article included 269 CAD patients and 131 healthy controls,and used an XGB model for prediction,with an area under the ROC curve of 0.728(95%CI 0.623-0.824).Further model shap explanations showed that BNP,left ventricular ejection fraction,homocysteine,and hemoglobin(p<0.001)were the main contributing factors.(3)Regarding multimodal data fusion,this study proposes a four-input convolutional neural network architecture that combines one-dimensional and two-dimensional convolutional neural networks to automatically extract and integrate features from baseline data,raw cardiac sound sequences,and cardiac sound signal time-frequency images.The results show that this method significantly improves the accuracy of coronary heart disease detection,with classification accuracy,sensitivity,and specificity reaching90.49%,94.78%,and 87.33%,respectively,outperforming the effect of diagnosing coronary heart disease using single-modality data.Compared with existing research,this method can effectively capture potential information in signals by leveraging the advantages of deep learning,providing more comprehensive and reliable diagnostic evidence for early non-invasive and non-destructive detection of coronary heart disease.Therefore,the contribution of this study lies in proposing a new multi-input convolutional neural network architecture and successfully applying it to non-invasive and non-destructive detection of coronary heart disease,improving the accuracy and reliability of coronary heart disease detection,and providing valuable references and references for future related research.(4)Based on the constructed coronary heart disease diagnosis model,design and implement a coronary heart disease auxiliary diagnosis system.According to the analysis of system requirements,design modules such as auxiliary diagnosis,case management,and statistical analysis,and use machine learning algorithms to automatically analyze and diagnose whether patients have coronary heart disease,providing decisionmaking support for doctors in auxiliary diagnosis.