Abstract
The elucidation of the brain’s anatomical and functional organisation during specific tasks is a challenging field in modern brain research. There is also a growing interest in the field of brain connectivity and its relation to specific motor and mental tasks, as well as neurodegenerative diseases like Parkinson’s and Alzheimer’s. In this thesis, a novel approach for modelling motor tasks is proposed. This approach combines diffusion adaptation and brain connectivity measures in order to build models which describe complex tasks through time and space. In particular, an S-transform based measure is introduced to estimate the connectivity on single-trial basis. The connectivity values, corresponding to different frequency bands across time, are effectively coupled with diffusion adaptation. The diffusion strategy exploits the time-space characteristics in a distributed and collaborated manner, and leads to an enhanced model for motor or mental tasks. Specifically, the imaginary part of S-transform coherency is introduced as an EEG connectivity measure. The performance improvement over the existing connectivity measures on a single-trial basis is demonstrated. Moreover, diffusion Kalman filtering is used as it performs well for nonstationary problems like this. This novel method is tested on various scenarios. Initially, its performance is demonstrated for simulated datasets which are based on realistic scenarios. Then, the method is applied to two datasets of real data. The first set of experiments includes a complex motor task of clockwise and anticlockwise hand movement and the second set includes a multi-modal dataset acquired from Parkinson’s patients. The results show that the connectivity enhanced modelling outperforms the simple case where connectivity information is ignored, and can build a robust task-related model.