UC San Diego

Cardiovascular diseases have been the leading cause of death over the world for the last two decades. In the United States, the cost for heart diseases is around $363 billion from year 2016 to 2017, including the consumption from health care services, medicines, and lost productivity due to death. To prevent heart diseases and further deterioration, real-time cardiovascular monitoring and early detection of abnormal cardiac activities are critical. With the advance of wearable sensor technology and maturity of machine learning models, researchers have gradually adopted such techniques in a great number of computing cardiology problems. This dissertation aims to continue the success through advancing signal processing techniques for wearable sensors and devising machine learning methods motivated by applications to cardiovascular problems.

To this end, we provide computational methodologies for heartbeat detection and three relevant applications: arrhythmia classification, biometric identification, and blood pressure estimation. Heartbeat detection is the primary step to complete the three aforementioned and a majority of cardiac signal applications; therefore, high quality heartbeat detection algorithm is required for promising outcome. Successful automated arrhythmia classification program can achieve early diagnosis of heart diseases and reduce the manual arrhythmia labeling workload. Cardiology-based biometric has attracted attention in the last few years due to its great privacy preserving property and cardiac activity indicator. Portable blood pressure monitor can substantially improve the healthcare environment since blood pressure is able to infer the cardiovascular functioning.

We organized this dissertation as follows. First, we exhibit high quality heartbeat detection algorithms for normal electrocardiograms (ECG), abnormal ECG, and seismocardiograms (SCG). In addition, we present Star-ECG, a ECG visualization tool for non-experts to easily identify heartbeats and compute heart rate variability. Next, we present competitive computer-aided arrhythmia diagnosis programs for both one-channel and 12-lead ECG. Subsequently, we demonstrate ECG and SCG biometric authentication models. Finally, we provide a wearable blood pressure monitoring approach that is capable of capturing a person's blood pressure during normal walking motion.