Deep Learning Algorithms for Time Series Analysis of Cardiovascular Monitoring Systems

Abstract

This thesis investigates and develops methods to enable ubiquitous monitoring of the most examined cardiovascular signs, blood pressure, and heart rate. Their continuous measurement can help improve health outcomes, such as the detection of hypertension, heart attack, or stroke, which are the leading causes of death and disability. Recent research into wearable blood pressure monitors sought predominately to utilise a hypothesised relationship with pulse transit time, relying on quasiperiodic pulse event extractions from photoplethysmography local signal characteristics and often used only a fraction of typically bivariate time series. This limitation has been addressed in this thesis by developing methods to acquire and utilise fused multivariate time series without the need for manual feature engineering by leveraging recent advances in data science and deep learning methods that showed great data analysis potential in other domains.