Estimation of working memory load using EEG signal processing

Cognitive load is the load imposed on the working memory during a cognitive process, and its measurement is of great importance in cognitive monitoring. This is to avoid mental overload of the user or interface and maintain efficiency and productivity in work performance, especially in critical or high mental load workplaces such as air traffic control, medical or emergency applications, and military operations. Among different techniques available to measure cognitive load, Electroencephalography (EEG) is known as the best physiological measure that can be applied in online and continuous measurement, detecting subtle variations, particularly in instantaneous load during cognitive processing. Notwithstanding the plethora of previous EEG-based systems for measuring various cognitive states and pathologies proposed to date, the estimation of cognitive load using EEG signals is still in its early stages, and there is still a need to validate the applicability of new approaches across different tasks, to establish the precision of cognitive load measurement, and investigate the underlying neural dynamics, when performing a cognitive task with varying difficulty levels.By this research, our main aim is to estimate cognitive load induced by cognitive tasks precisely, using EEG signals. Therefore, we firstly design two cognitive tasks to create new EEG databases: a silent reading task in three levels of difficulty and an arithmetic task in seven levels of difficulty. Previous empirical studies of discrete load levels in similar contexts to date have used at most five levels. Herein, we advance this to seven induced load levels, to test the existing limits of precision of EEG in discriminating fine task load levels. In previous EEG research, mainly spectral features have been deployed for this purpose, however to date; other features that could potentially provide more information on the varying task difficulty on the memory load have not been investigated. Further, in these studies the theta and alpha EEG frequency bands have been mainly investigated. Therefore, we secondly examine and compare the suitability of different classes of time, frequency and time-scale and information/entropy based features for discriminating induced cognitive loads, towards extending the existing knowledge of which features, channels and frequency bands are the most suitable for an EEG-based load classification system. As a result, we propose a set of features, including energy and entropy of the wavelet coefficients, as the best performing features with a high classification rate for discriminating the fine memory load levels. Our results demonstrate that the frontal lobe of the brain reflects the largest number of channels contributing to the task load discrimination for the extracted features and the delta frequency band is the most discriminative band. This suggests that a smaller number of EEG channels in one frequency band, may be needed for future similar work. This is while the delta frequency band has often been disregarded in many previous cognitive studies. Since, the effects of task difficulty on the brain dynamics are not well-understood, we thirdly investigate the relationship between the task difficulty and brain cortical activation towards an understanding of the underlying neural dynamics. This is pursued by means of variations of dynamical features such as correlation dimension, Hurst exponent, and entropy-based features in accordance with the level of task difficulty imposed. For instance, as the extracted correlation dimension values tended to increase as the task difficulty increased; indicating that the brain activity dimension/complexity increases with the increase of cognitive activity load or difficulty. Further, a decreased value of entropy-based features with increased task difficulty implies higher predictability and more regularity/order in the brain activity, while the decline in the extracted Hurst exponent values as the task load increased demonstrates that random behaviour of the signal decreases as the task load increases. The last two set of features may indicate the brain behaves in a more regular and focused manner when performing more difficult tasks. These results can confirm the changing dynamics of the brain when performing a cognitive task with different load or difficulty levels. In contrast, previous mental task studies have shown lower dimension or complexity and higher irregularity when the brain goes to a passive state or a state of relaxation, compared to a cognitive state.On the other hand, this change can be associated with the dynamical processes of the underlying neural ensembles such as; complexity and synchronization between the EEG channels; we therefore investigate these two dynamics to examine their relationship with the task difficulty. The results of our entropy-based complexity measures indicate that with increasing task difficulty, not only does the disorder of the signals decline but also the complexity grows. The growing complexity can also reflect the higher activation of neural ensembles involved, as the task difficulty increases. The investigation of synchronization between EEG channels using phase locking values also demonstrates more synchronization among the channels as the task difficulty increases. The investigation of these dynamical aspects of the cognitive task provide more information into the brain functions, which has not been investigated in the case of task difficulty variations, previously.We also present classification results of a feature set with high accuracy of 98% across twelve subjects for the selected frontal channels across the seven load levels, in a multi-channel and subject-independent manner. This is while the highest classification accuracy previously reported in a similar context is 88% for four load levels across eight subjects. These findings extend prior work and bring us a step closer to the use of precise EEG-based technologies in BCI and HCI research, by introducing more suitable features, channels and frequency band for fine discrimination of cognitive loads with high accuracy, in similar contexts and correlates the difficulty a cognitive task to the relevant underlying brain operations, to provide more insight into the brain functions