Generating Synthetic Ballistocardiogram Signals with Custom Respiration Patterns

Ballistocardiography is a valuable tool to noninvasively monitor patients in everyday situations. As a Ballistocardiogram (BCG) signal depicts mass shifts in the human body, it picks up the respiratory process which holds valuable information for medical staff. Being able to detect disordered breathing patterns in such a signal would significantly increase the effectiveness of BCG monitoring. However, to train a classifier with such capabilities a labeled dataset of respiration and BCG signals from affected patients would be needed. As there is currently no such dataset available, we propose a method to generate different respiration patterns that can be concatenated without interruptions or interference to create a continuous respiration wave. These waves can then be used to create BCG signals that aims to emulate signals of patients with different breathing patterns. Our respiration signal generator produces waves close to recorded ones. Therefore, the resulting BCG signals should be suitable to create training sets for classifiers that identify respiration patterns in BCG signals or to augment existing training data.