Machine Learning Assisted Automatic Annotation of Isovolumic Movement and Aortic Valve Closure using Seismocardiogram Signals

Recently, Seismocardiogram (SCG) is used to monitoring vital cardiac health parameters through relevant SCG peaks annotation. However, accurate annotation of the SCG peaks is a challenging task. In this paper, we design binary classifiers to annotate the Isovolumic movement and the Aortic valve closure events of SCG. Five binary classifiers, such as Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, and Gaussian Nave Bayes, are implemented atop customized features "Amplitude" and "Time of Appearance". The experiments are carried out on the "cebs" database available at Physionet public repository. In particular, seven rigorous performance metrics are employed to evaluate the classifiers’ annotation ability, followed by the 5-fold cross-validation to ensure the classifiers’ performance reliability. Experiment results show that RF consistently performs better for both IM and AC annotation. On the contrary, SVM and LR perform poorly for IM and AC, respectively.