Identification of respiratory phases using seismocardiogram: A machine learning approach

This study was aimed at developing an algorithm that could identify the respiratory phases, i.e. inspiration (I) or expiration (E), by analysing seismocardiogram (SCG) cycles. In order to better assess SCG cycles, it is needed to discriminate the cycles based on their position in the respiratory phases. The total 2146 SCG cycles obtained from 45 subjects were studied, in which 1109 cycles were in phase I, and the rest in phase E. Support vector machine (SVM), a powerful machine learning algorithm, was employed to identify the respiratory phase of SCG cycles. The systolic interval of each SCG cycle was divided to 32 equal bins, and the averages of these bins obtained the feature vector associated with each cycle. The SVM model was trained using half the data, and then was tested on the other half. The developed model could correctly identify 88% of the testing data. The obtained results are promising and can establish a solid ground for further analysis.