Bed-based ballistocardiography: physiological assessment and system design
Abstract
The ballistocardiogram (BCG) is an information-rich bio-signal that can give researchers an additional perspective on the complex dynamics of the cardiac cycle. As a relevant alternative to the electrocardiogram (ECG), the BCG represents the transduced force produced by the ejection of blood from the heart around the aortic arch. Due to its characteristic propagation through the participant’s body and corresponding measurement system, the BCG offers a combination of physiological information about the participant. In its current state of research, integrated BCG systems are being developed to predict various physiological health measures. This thesis seeks to further develop the range of physiological predictions available to BCG, focusing on Kansas State University’s bed-based BCG acquisition system. Specific physiological prediction of instantaneous blood pressure is first investigated in this thesis. Utilizing parallel BCG sensors integrated into a bed, frequency-domain analysis is used to predict instantaneous blood pressure during simulated apnea events. With 16/20 participants resulting in strongly correlated predicted blood pressure and an average mean absolute error (MAE) across all participants less than 5 mmHg, this frequency analysis method could offer improved morphology-independent blood pressure prediction using BCG. Preliminary classification of apnea events is also performed for a single extended run (~10 minutes) of simulated apnea and rest events. With an average of 95% balanced accuracy across six folds, further investigation on a larger dataset should be performed to follow up on this preliminary connection between beat-by-beat apnea classification and the BCG frequency domain. This analysis seeks to expand the use case of the BCG as an “all-in-one” apnea detection signal. The frequency domain is chosen for investigation due to a key limitation often observed in BCG research. System-level differences and participant-specific body positions can lead to morphological changes between participants. Time-domain analysis is often limited to individual systems/participants due to a combination of factors such as system design, sensor coupling, and cardiac output. With the intuition that the bed-body system can be characterized and has an effect on the BCG’s morphology, the next logical step in the included research is to identify if the BCG can be used to predict relevant participant-specific characteristics. Using DEXA as a gold standard to measure regional and total body fat percentage, the BCG morphology is investigated to identify any features relevant to body composition. Through the use of a feature search of common BCG signal characteristics, a baseline improvement from BMI-only body composition prediction can be observed. The inclusion of BCG features in BMI-based body composition estimation provides preliminary evidence that BMI-based body composition can be improved. BCG could offer improved predictive performance, similar to current body composition estimates, using a combination of predictors (BMI, bioimpedance, anthropometric measurements, etc.). After discussing the relevant analysis performed on collected BCG data, this thesis concludes with design and analysis documentation relevant to Kansas State University’s bed-based BCG acquisition system. This documentation includes design history and implementation details that are recommended for future BCG system modifications. To expand future Kansas State research pathways, the BCG system at Kansas State should be further improved with the end goal of data collection related to the preliminary investigations discussed in this thesis.