Objectives: The weight and volume of battery-powered wireless electroencephalography
(EEG) systems are dominated by the batteries. Battery dimensions are in
turn determined by the required energy capacity, which is derived from the system
power consumption and required monitoring time. Data reduction may be carried
out to reduce the amount of data transmitted and thus proportionally reduce
the power consumption of the wireless transmitter, which dominates system power
consumption. This thesis presents two new data selection algorithms that, in addition
to achieving data reduction, also select EEG containing epileptic seizures and
spikes that are important in diagnosis.
Methods: The algorithms analyse short EEG sections, during monitoring, to
determine the presence of candidate seizures or spikes. Phase information from
different frequency components of the signal are used to detect spikes. For seizure
detection, frequencies below 10 Hz are investigated for a relative increase in frequency
and/or amplitude.
Significant attention has also been given to metrics in order to accurately evaluate
the performance of these algorithms for practical use in the proposed system.
Additionally, signal processing techniques to emphasize seizures within the EEG
and techniques to correct for broad-level amplitude variation in the EEG have been
investigated.
Results: The spike detection algorithm detected 80% of spikes whilst achieving
50% data reduction, when tested on 992 spikes from 105 hours of 10-channel scalp
EEG data obtained from 25 adults. The seizure detection algorithm identified 94%
of seizures selecting 80% of their duration for transmission and achieving 79% data
reduction. It was tested on 34 seizures with a total duration of 4158 s in a database
of over 168 hours of 16-channel scalp EEG obtained from 21 adults. These algorithms
show great potential for longer monitoring times from miniaturised wireless
EEG systems that would improve electroclinical diagnosis of patients.