Research On Heart Function Grading Of Heart Failure Based On Multimodal Data

Heart failure is associated with high incidence and mortality rates,leading to a significant burden on patients’ lives.Diagnosing and treating heart failure is a significant challenge for the healthcare system,requiring the early detection of heart failure and the assessment of its severity.Developing a heart failure diagnosis model and heart function grading model is of great social significance and research value.In the guidelines for the prevention,detection,and management of heart failure,the diagnostic process involves electrocardiogram,echocardiography,biochemical examination,and other routine examinations,encompassing multi-modal data.However,most existing research utilizing machine learning for heart failure diagnosis or heart function grading is based on single-modal data,overlooking the complementary information available from multiple modal data sources.Consequently,this paper proposes a multimodal data fusion method,utilizing electrocardiogram,echocardiogram,and biochemical examination data to diagnose heart failure and study heart function grading,for the first time.The paper’s primary focus includes:Firstly,In this study,two multi-modal datasets were constructed.Following guidelines for heart failure diagnosis,4190 heart failure-related medical records and 1976 heart function grading-related medical records were extracted from the hospital’s medical records.From each medical record,the requisite electrocardiogram,echocardiography report,and biochemical examination data were extracted.Given the considerable noise in the electrocardiogram data in this paper,noise removal processing was executed to mitigate noise in the electrocardiogram data.Moreover,data preprocessing techniques such as missing value filling and normalization were applied to the echocardiography report and biochemical examination data.Secondly,Multi-modal data representation and fusion methods were studied in this research,with a proposed fusion method for electrocardiogram data,echocardiography report data,and biochemical examination data.To extract features from the fused multi-modal data,a feature extraction neural network module based on residual and convolutional networks was constructed.Following this,a heart failure diagnosis model was constructed,utilizing the multimodal data fusion method and feature extraction neural network module.In order to construct the heart function grading model,deep learning and statistical machine learning methodologies were combined.Finally,The heart failure diagnosis and heart function grading experiments were conducted in this study.The proposed heart failure diagnosis model was validated by achieving a 93.95%final diagnostic accuracy,and AUC of 0.98.In the heart function grading experiment,an accuracy of 91.20% and an F1 value of 91.11% were achieved,A 7.6% improvement in accuracy was achieved,compared to using only deep learning.Moreover,ablation experiments were undertaken,which revealed that multi-modal data-based heart failure diagnosis and heart function grading experiments greatly improved various evaluation indicators compared to single-modal data.