Dr. Tingting Zhu, Senior Researcher, Department of Engineering Sciences, University of Oxford, UK

Speech Title: Personalised Patient Physiology Monitoring using Bayesian Hierarchical Gaussian Processes

Abstract: With the rapid increase in volume of wearable devices, automated algorithms are employed to label physiological time-series data. The management of chronic diseases can be performed by monitoring continuous time-series vital-sign data (such as blood pressure and heart rate) via low-cost wearable devices. Automated algorithms may then be used with the resulting data to provide early warning of deterioration of the health of an individual. Such algorithms are typically trained for a large population without taking into account the time-variability and inter-subject variability of the data being collected. Here, we propose using Bayesian Gaussian processes to model the changes of vital signs over time for a patient undergoing continuous monitoring. Furthermore, we use Hierarchical Gaussian processes to infer the hidden latent structure of vital sign trajectory for each individual patient or group of patients who sharing similar patterns. In order to identify any patterns that correspond to “normal” or “abnormal” physiology, we further propose using the symmetric Kullback-Leibler divergence of multivariate normal distributions to provide a metric for identifying “abnormal” patterns from a model of “normal” latent trajectories. We have demonstrated our approach using two real datasets for different monitoring scenarios. Our model is compared to the performance of state-of-the-art unsupervised clustering algorithms, as well as two one-class classifiers, which demonstrates the superiority of the proposed method in identifying abnormal physiological episodes.

Keywords: wearable sensors, personalised medicine, patient monitoring