Auxiliary Diagnosis of Acute Heart Failure Based on Ballistocardiography and Electronic Health Records

Acute heart failure (AHF) is characterized by its sudden onset and urgency, highlighting the critical importance of home-based monitoring in the timely detection of disease and the securing of treatment time. This study aims to develop a novel early warning method for AHF by integrating models based on ballistocardiography (BCG) signals and electronic health records (EHR). A total of 64 heart failure patients were enrolled, including 20 patients with AHF and 44 patients with chronic heart failure. First, piezoelectric sensors were placed under the subjects' pillows to collect BCG signals while they were lying. Subsequently, EHR information of the subjects, such as height and smoking history, was meticulously recorded. Finally, the output probabilities of the BCG-based and EHR-based classification models were integrated and fed into a random forest for the classification of AHF. Under the leave-one-subject-out validation, the proposed signal fusion model demonstrated superior classification performance compared to individual signal-based early warning methods, achieving an accuracy of 90.62%, sensitivity of 80%, and specificity of 95.45%. Notably, the high-frequency information of BCG signals and the presence of congenital heart disease in the EHR data play an important role in the early warning of AHF. The integration of BCG signals and EHR can significantly improve the diagnostic accuracy of AHF, which is of great significance for future home-based early warning of AHF.