Thesis defense Jeroen Geuze (Donders Series 160)

27 May 2014

Promotor: Prof.dr. P. Desain, copromotor: dr. J.D.R. Farquhar

Brain Computer Interfaces for Communication: Moving beyond the visual speller

This thesis is about Brain Computer Interfaces (BCIs) for Communication, specifically on how they can be improved. A BCI is a system that allows someone to control a computer by using only their brain activity. One of the best known BCIs for communication is the visual speller, which has first been developed in 1988, and has been researched intensively since then.

Improvement of the performance of an existing system, the visual speller, was investigated. By combining different manipulations of the stimuli used in the visual speller, the performance was increased. However, the visual speller was still dependent on gaze direction.

Some patient groups that would benefit from a communication BCI are not able to direct their eye gaze anymore. Therefore, a tactile speller was developed to improve the match of the BCI with the target group (patients). The results were compared with existing visual speller paradigms in terms of classification performance and elicited ERPs. It was shown that it is possible to use a tactile speller for communication. The tactile speller provides a useful alternative to the visual speller, especially for people whose eye gaze is impaired.

The next improvement was to move towards more natural communication. First semantic priming was investigated. Semantic Priming could be used to traverse a semantic network and determine which word a person has in mind. By using machine learning techniques it is possible to analyse and classify short traces of brain activity, which could, for example, be used to build a Brain Computer Interface (BCI). It was shown that semantic priming can be detected significantly above chance level for all subjects.

Next, A BCI based on semantic relations was investigated. The BCI determines which word a subject has in mind by presenting probe words using an intelligent algorithm. Subjects indicate when a presented probe word is related to the word they have in mind by a single finger tap. The detection of the neural signal associated with this movement is used to decode the target word. It was shown that an intelligent algorithm used to present the probe words has a significantly higher performance than a random selection of probes. Simulations demonstrate that the BCI also works with larger vocabulary sizes, and the performance scales logarithmically with vocabulary size.

This thesis has shown some possible improvements, by using better stimulus encoding, using more appropriate stimulus modalities or trying to detect higher level cognitive concepts. Whilst all of these approaches show some promise, none was the 'silver bullet' which would make useable BCIs a reality. In fact whilst it is clear that patient BCIs will be useful in the next few years, it is also clear that much more work is needed before they represent a viable alternative communication modality for the majority of patients and can be moved out of the lab and into the homes of users in need.