Identification of audio evoked response potentials in ambulatory EEG data

Alfahad, Othman Abdulaziz

Abstract

Electroencephalography (EEG) is commonly used for observing brain function over a period of time. It employs a set of invasive electrodes on the scalp to measure the electrical activity of the brain. EEG is mainly used by researchers and clinicians to study the brain’s responses to a specific stimulus - the event-related potentials (ERPs). Different types of undesirable signals, which are known as artefacts, contaminate the EEG signal. EEG and ERP signals are very small (in the order of microvolts); they are often obscured by artefacts with much larger amplitudes in the order of millivolts. This greatly increases the difficulty of interpreting EEG and ERP signals.
Typically, ERPs are observed by averaging EEG measurements made with many repetitions of the stimulus. The average may require many tens of repetitions before the ERP signal can be observed with any confidence. This greatly limits the study and use
of ERPs. This project explores more sophisticated methods of ERP estimation from measured EEGs. An Optimal Weighted Mean (OWM) method is developed that forms a weighted average to maximise the signal to noise ratio in the mean. This is developed
further into a Bayesian Optimal Combining (BOC) method where the information in repetitions of ERP measures is combined to provide a sequence of ERP estimations with monotonically decreasing uncertainty. A Principal Component Analysis (PCA) is
performed to identify the basis of signals that explains the greatest amount of ERP variation. Projecting measured EEG signals onto this basis greatly reduces the noise in measured ERPs. The PCA filtering can be followed by OWM or BOC. Finally, cross
channel information can be used. The ERP signal is measured on many electrodes simultaneously and an improved estimate can be formed by combining electrode measurements. A MAP estimate, phrased in terms of Kalman Filtering, is developed using all electrode measurements.
The methods developed in this project have been evaluated using both synthetic and measured EEG data. A synthetic, multi-channel ERP simulator has been developed specifically for this project.
Numerical experiments on synthetic ERP data showed that Bayesian Optimal Combining of trial data filtered using a combination of PCA projection and Kalman Filtering, yielded the best estimates of the underlying ERP signal. This method has been applied to subsets of real Ambulatory Electroencephalography (AEEG) data, recorded while participants performed a range of activities in different environments. From this analysis, the number of trials that need to be collected to observe the P300 amplitude and delay has been calculated for a range of scenarios.

Citation

Alfahad, O. A. (2018). Identification of audio evoked response potentials in ambulatory EEG data. (Thesis). University of Hull. Retrieved from https://hull-repository.worktribe.com/output/4398611

• Thesis Type: Thesis
• Deposit Date: Sep 25, 2023
• Publicly Available Date: Sep 25, 2023
• Keywords: Medical engineering
• Public URL: https://hull-repository.worktribe.com/output/4398611
• Additional Information: Department of Medical Engineering; University of Hull
• Award Date: Sep 1, 2018