Methods for the Classification of Biosignals Applied to the Detection of Epileptiform Waveforms and to the Recognition of Physical Activity [Menetelmiä biosignaalien luokitteluun sovellettuna epileptiformisten aaltojen havaitsemiseen ja fyysisen aktiviteetin tunnistamiseen]

Biosignals are such signals that quantify the physiological processes of a living organism. Classification of biosignals aims at inferring the physiological condition of the organism based on the biosignals obtained from it. In this thesis, the classifications of two biosignals originating from the human body are studied in detail: the electroencephalogram (EEG) and acceleration signals recorded from body-worn sensors (body accelerometry). EEG quantifies the electrical activity of the brain. In this thesis, EEG recorded in hospital operating room and intensive care unit environments is classified to detect epileptiform brain activity which is a potentially brain-damaging phenomenon. Wavelet subband entropy of EEG is shown to be statistically associated with epileptiform activity both in operating room patients under sevoflurane-induced anesthesia and in intensive care unit patients resuscitated after cardiac arrest. The results support the hypothesis that epileptiform activity can be continuously monitored in both clinical settings. Body accelerometry quantifies the movements of the human body with bodyworn sensors. In this thesis, body accelerometry is classified for activity recognition purposes, i.e. the purpose is to detect the type of physical activity of the subject from the body acceleration signals. State-of-the-art offline classification results are obtained in two studies. In addition, conversion of the presented offline activity classification algorithms to an online version is demonstrated. The results confirm that multiple classes of daily physical activities and sports can be reliably recognized with body accelerometry.