|M.Sc Student||Shoor Ehud|
|Subject||Detection and Analysis of Movement-Related Potentials|
for Brain-Computer Interface Applications
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus Gideon Inbar (Deceased)|
|Full Thesis text|
Brain-Computer-Interface (BCI) is a direct, non-muscular, control channel between the human brain and a computer or other external devices. One method for operating BCI is to utilize movement-related brain potentials (MRP's), which are the electro-encephalographic (EEG) components related to voluntary movements. The main problem in developing MRP based BCI is reliably detecting single-trial MRP's from the ongoing EEG. The challenge stems from the extremely unfavorable signal-to-noise ratio (SNR) at which the MRP's are recorded, where the background brain activity is referred to as noise. Furthermore, both the desired response and the background noises are non-stationary signals and in general, have similar spectral characteristics. The research focuses on the analysis and detection of MRP's and contains two parts: the first part follows-up a past study of movement related potentials in the frequency domain. The past paradigm was refined from being auditory cue driven to self-paced setup, where the subjects initiate their movements in a voluntary, asynchronous manner. The new setup better emulates natural usage-model and also allows avoiding potential artifact brain responses which are associated to the external auditory cues and not to the movement intentions. In the second part, a novel MRP detection method based on spatial de-noising algorithm was developed and validated. The proposed algorithm is directly addressing the fundamental MRP detection challenge, which stems from the poor signal-to-noise ratio in which the signals are being recorded. The developed method is based on spatial weighted-least-squares (SWLS) approach, which utilizes the spatial correlation between different recording electrodes over the scalp, and provides enhanced MRP detection capabilities. One of the main benefits of the proposed method is its adequate robustness for operation over relatively short time intervals, without assuming any a priori constraints. This property is highly important for EEG processing because of the non-stationary nature of the signals. Another advantage of the developed algorithm is its high level of flexibility, allowing representing the relation between different recordings by FIR filter rather than just a proportional factor, which enables to tolerate propagation delays between the sources and the electrodes and provides fine matching resolution. The algorithm was tested and operated well in asynchronous detection tasks, and seems as a promising direction for a robust MRP based brain-computer-interface system.