Digital Biomarker Discovery for Non-Invasive Health Monitoring with Acoustic and Vibration Signals

Thumbnail Image
Semiz, Beren
Inan, Omer T.
Associated Organizations
Supplementary to
A biomarker is a parameter that can be used to objectively quantify a physiological or pathophysiological process. Biomarkers used in medicine are commonly derived from blood, saliva, urine or other bodily fluids, and in many cases are used to inform medical decisions. There is a new emerging class of biomarkers, digital biomarkers, which are measures collected through connected digital tools, generally across multiple hardware and software layers. This work describes the use of wearable acoustic and vibration measurements to derive digital biomarkers, which can be used together with existing medical information to assist in clinical decisions. Acoustic and vibration signals carry information that is in many cases complementary to electrophysiology or movement, but the signals are not fundamentally well understood. This leads to a limited one-to-one correspondence between signal characteristics and important health parameters. To that end, this work aims to investigate this correspondence through signal processing, data-driven feature discovery and statistical techniques for deriving accurate and clinically relevant digital biomarkers. In addition, acoustic and vibration signals exhibit substantial inter-subject and intra-subject variability, thus their use in classical diagnostic approaches have not been successful in the past. Rather than focusing on adapting these signals as diagnostic tools, this work aims to derive and employ new algorithms to detect and track the relative changes in health, e.g. exacerbation in clinical state and/or response to a specific treatment, for a given subject over time. The first part of this dissertation discusses how wearable acoustic measurements can be leveraged in biomechanics, specifically in joint health assessment, to derive clinically useful digital biomarkers. The first work presents the use of knee acoustical emissions captured through miniature sensors to derive a clinically relevant joint health score for the evaluation of juvenile idiopathic arthritis. Then, a novel click detection and classification algorithm leveraging the Teager Energy Operator is presented to detect the clicks in joint sound signals and distinguish between physiologic and pathologic clicks. The second part of this dissertation studies the wearable acoustic and vibration measurements for cardiovascular assessment in two different applications. The first application involves pump thrombosis detection in left ventricular assist devices based on analyzing the operating sounds of the pump with machine learning algorithms. The second application is the non-invasive estimation of stroke volume based on wearable seismocardiogram and phonocardiogram measurements taken from the sternum. Overall, this dissertation presents novel frameworks leveraging wearable vibration and acoustic measurements for knee joint and cardiovascular health assessment. Once verified and validated through large studies, such systems can potentially assist in clinical decisions and improve the management of various diseases and injuries outside the physical confines of the clinic.
Date Issued
Resource Type
Resource Subtype
Rights Statement
Rights URI