Research on Preprocessing and Analysis Methods for Ballistocardiogram Signals

Arrhythmia is caused by abnormal electrical activity in the heart, with atrial fibrillation (AF) being one of the most common types. Ballistocardiography (BCG), a method capable of unobtrusively detecting cardiac activity, is particularly suitable for at-home screening and monitoring of AF. This study aims to identify AF through the analysis of BCG signal features, focusing on three main aspects: morphological, statistical, and signal analysis characteristics. Template modeling and matching methods were employed to detect heartbeats and J-peaks in BCG signals, from which the inter-beat intervals were calculated. Features were extracted by analyzing the variance and range of these intervals. Additionally, higher-order statistical measures were applied to assess the amplitude distribution of BCG signals before and after energy transformation. Using these extracted features, a support vector machine (SVM) classifier was utilized to distinguish AF from non-AF signals. The classifier's performance was evaluated through 10-fold cross-validation, achieving an accuracy of 91.8%, sensitivity of 92.51%, specificity of 91.07%, and precision of 91.26%.