Sensor fusion of physiological signals for mobile mental state monitoring.

Abstract

Coupled with wide connectivity, abundant data, and growing computational power, the development of body sensor networks for continuous mobile monitoring of physiological signals has made breakthroughs in the advancement of e-Health systems in clinical and non-clinical applications. However, translating physiological signals to useful health information face key challenges; these originate from the nonstationary, nonlinear, subject-dependent, and time-varying characteristics of the measured physiological signals. Such challenges are reflected in system performance metrics by a decrease in sensitivity or an increase in the false alarm rate. While a trade-off exists between these two metrics, several techniques have been proposed to reduce the false alarm rate while still maintaining a relatively high sensitivity. As one of the potentially effective approaches, sensor fusion of different signal sources at the data, feature, and decision levels has been employed in detection and estimation schemes for health monitoring systems to guarantee more robust and reliable health assessments from the acquired signals. This thesis aims at incorporating sensor fusion methodologies for estimating measures of mental health in two applications: First, in the non-clinical application of estimating a cognitive state (i.e., a level of mindfulness or focused awareness) during meditation practices using ECG and EEG signals, and second, in the clinical application of detecting status epilepticus (i.e., recurrent or prolonged seizures) in coma patients using multi-channel EEG signals. To this end, the thesis proposes: 1) application-specific features of ECG and EEG signals, 2) methods for generating local (per signal source) estimates decisions of the target variable (i.e., mindfulness level seizure detection decisions), and 3) a knowledge-based heuristic fusion method for combining local decisions into a more robust global decision. In order to test the performance of the proposed methodologies in the meditation application, EEG and EC