Title:  Processing and Analysis of the Seismocardiogram to Enable Estimations of Blood Volume Decompensation Status

Committee:

Dr. Omer Inan, ECE, Chair, Advisor

Dr. Ying Zhang, ECE

Dr. Rishikesan Kamaleswaran, Emory

Dr. Michael Sawka, BioSciences

Dr. Jin-Oh Hahn, Univ of Maryland

Abstract:

Title:  Processing and Analysis of the Seismocardiogram to Enable Estimations of Blood Volume Decompensation Status

Committee:

Dr. Omer Inan, ECE, Chair, Advisor

Dr. Ying Zhang, ECE

Dr. Rishikesan Kamaleswaran, Emory

Dr. Michael Sawka, BioSciences

Dr. Jin-Oh Hahn, Univ of Maryland

Abstract: The objective of this research was to develop a system comprised of wearable sensing and machine learning algorithms to continuously estimate an individual's hypovolemic or blood volume status. Hypovolemia is a leading cause of preventable death, with many potentially overlapping causes occurring in both hospital and field locations. The objective of this work is to explore whether a multi-modal wearable system comprised of noninvasive electro-mechanical cardiac sensors paired with machine learning is sufficiently capable of estimating an individual's blood volume status. To this end, an intensive large animal study was carried out in which noninvasive signals and catheter pressure waveforms were recorded as the animals underwent changes in relative and absolute blood volume. Features were extracted from both the noninvasive and catheter signals and compared against each other for quality. These features were used to train a model to predict individual-specific hypovolemic status during hemorrhage. This model was later expanded to include training data from the entire protocol to predict hypovolemic status during relative and absolute hypovolemia as well as during resuscitation. The features derived from the seismocardiogram were determined to be key in estimating hypovolemic status. The processing of this signal is still in the early stages, subject to many types of noise and requiring human oversight. The remainder of this work presents three algorithms developed for reliably processing seismocardiogram signals and indicates future directions for research.