Research On Multimodal Fusion Algorithm For Cardiomyopathy Diagnosis

In recent years,with the progress of technology and the development of the field of artificial intelligence,the use of artificial intelligence in the medical field has become increasingly widespread,and the use of machine learning methods for disease diagnosis is also becoming more common.Some researchers have achieved some success in using machine learning methods to diagnose heart diseases by utilizing data such as electrocardiogram and echocardiogram.However,currently these methods still face challenges such as incomplete diagnosis categories,low performance in partial diagnosis classification,and inability to meet practical application scenarios by using single-modal data.These methods train models using single-modal data,leading to incomplete feature acquisition by the models.Therefore,this thesis proposes a multimodal fusion model that uses electrocardiogram data,echocardiogram reports,and biochemical examination data for multimodal fusion.On one hand,this model solves the problem of incomplete diagnosis categories and enables the diagnosis of multiple types of myocardial diseases.On the other hand,by utilizing multimodal data,the model improves the diagnostic classification performance in practical application scenarios.This thesis investigates the multimodal fusion algorithm and proposes a multimodal fusion algorithm for myocardial disease diagnosis by combining electrocardiogram(ECG)data,echocardiography reports,and biochemical test data.A multimodal fusion model is constructed and trained to learn the features of different modalities,thereby improving the diagnostic classification performance of the model.To train the multimodal fusion model,a multimodal dataset is constructed by extracting and preprocessing ECG data,echocardiography reports,and biochemical test data from the raw data files,and selecting the required data for this study.Based on the aforementioned multimodal data,a multimodal fusion model composed of a feature extractor,a feature fusion module,and a classifier was proposed.The model addresses the issue of low diagnostic classification performance of previous models due to partial use of data and inability to achieve expected results in practical applications by integrating features from different modalities.The proposed multimodal fusion model in this study achieved an accuracy of 89.87%,recall of 91.20%,F1 score of 89.13%,and an overall accuracy of 89.72% for the diagnosis of myocardial disease.Furthermore,the model was compared to baseline models through comparative experiments and ablation studies,and demonstrated superior performance,thereby confirming the effectiveness of the proposed multimodal fusion model.At the end of this thesis,the proposed multimodal fusion model and dataset construction method will be applied to create a myocardial disease diagnosis system.This system is constructed based on the client-server architecture and includes functions for diagnosing and classifying myocardial disease,as well as extracting medical records data.In practical use,users can input their medical history and electrocardiogram data to obtain a diagnosis of their medical condition,as well as results of their echocardiography report and biochemical analysis data.