Applications of non-invasive brain-computer interfaces for communication and affect recognition



Journal Title

Journal ISSN

Volume Title



Various assistive technologies are available for people with communication disorders. While these technologies are quite useful for moderate to severe movement impairments, certain progressive diseases can cause a total locked-in state (TLIS). These conditions include amyotrophic lateral sclerosis (ALS), neuromuscular disease (NMD), and several other disorders that can cause impairment between the neural pathways and the muscles. For people in a locked-in state (LIS), brain-computer interfaces (BCIs) may be the only possible solution. BCIs could help to restore communication to these people, with the help of external devices and neural recordings.

The present dissertation investigates the role of latency jitter on BCIs system performance and, at the same time, the possibility of affect recognition using BCIs. BCIs that can recognize human affect are referred to as affective brain-computer interfaces (aBCIs). These aBCIs are a relatively new area of research in affective computing. Estimation of affective states can improve human-computer interaction as well as improve the care of people with severe disabilities. The present work used a publicly available dataset as well as a dataset collected at the Brain and Body Sensing Lab at K-State to assess the effectiveness of EEG recordings in recognizing affective states.

This work proposed an extended classifier-based latency estimation (CBLE) method using sparse autoencoders (SAE) to investigate the role of latency jitter on BCI system performance. The recent emergence of autoencoders motivated the present work to develop an SAE based CBLE method. Here, the newly-developed SAE-based CBLE method is applied to a newly-collected dataset. Results from our data showed a significant (p < 0.001) negative correlation between BCI accuracy and estimated latency jitter. Furthermore, the SAE-based CBLE method is also able to predict BCI accuracy.

In the aBCI-related investigation, this work explored the effectiveness of different features extracted from EEG to identify the affect of a user who was experiencing affective stimuli. Furthermore, this dissertation reviewed articles that used the Database for Emotion Analysis Using Physiological Signals (DEAP) (i.e., a publicly available affective database) and found that a significant number of studies did not consider the presence of the class imbalance in the dataset. Failing to consider class imbalance creates misleading results. Furthermore, ignoring class imbalance makes comparing results between studies impossible, since different datasets will have different class imbalances. Class imbalance also shifts the chance level. Hence, it is vital to consider class bias while determining if the results are above chance. This dissertation suggests the use of balanced accuracy as a performance metric and its posterior distribution for computing confidence intervals to account for the effect of class imbalance.



Brain-computer interfaces (BCIs), Electroencephalogram (EEG), Classification, P300 speller, Emotion recognition, Machine learning

Graduation Month



Doctor of Philosophy


Department of Electrical and Computer Engineering

Major Professor

David E. Thompson