Classification of physiological sensor signals using artificial neural networks

Abstract

Physiological signals have certain prominent characteristics that distinguish them from other types of physiological signals which are familiar to experts and assessed by inspection. The aim of this paper is to develop a computational model that can distinguish electrocardiogram, galvanic skin response and blood pressure signals acquired from sensors as well as detect corrupted signals which can arise due to hardware problems including sensor malfunction. Our work also investigates the impact of the signal modeling for various time lengths and determines an optimal signal time length for classification. This provides a method for automatic detection of corrupted signals during signal data collection which can be incorporated as a support tool during real-time sensor data acquisition. © Springer-Verlag 2013.