Available Projects
On the development on Blind Source Separation techniques with sparse signals: application to brain signal recordings
Blind source separation (BSS) is a mathematical technique that allows one to "unmix" a set of mixed signals in such a way that makes minimal (yet powerful) assumptions and yet results in outputs where the unmixed signals represent signals of interest which are (usually) buried deep in noisy signals. Brain signals are no different - successfully de-noising brain electric signals is an ongoing challenge in research.
Primary supervisor: Professor Christopher James - Email: c.james@warwick.ac.uk
Project detail:
BSS techniques such as Independent Component Analysis (ICA) have been developed over the years to successfully unmix mixed signal recordings. For the most part the methods have met with success - albeit usually under strict conditions. The electroencephalogram (EEG) - the recording of the electrical activity of the brain from the scalp - fits this situation well, where multiple underlying brain sources are "mixed" as measured over a set of electrodes. Standard ICA techniques show much promise in the EEG in many situations but tend lose efficacy as the number of recording channels is reduced, and simply cannot work, out-of-the-box, on a single recording channel. As using fewer recording channels in the EEG has many benefits - especially in situations where using EEG away from the clinic / in the home or community - we have devised methodologies that allow us to apply ICA techniques to single channel recordings to good effect.
Single channel ICA methods alone, however, are not the answer as they lack "spatial" awarenesss - where something is happening in the brain is important as when and what that activity looks like, for this reason we have developed techniques that successfully apply ICA to "few" recording channels - this has the benefit of needing only a few recording channels but still has the spatial information that is crucial in EEG signal analysis.
This project concerns the continued development of the so-called "space-time ICA" (stICA) that has been developed in my lab. As it stands the methods shows great promised but is still a) heavily dependent on manual intervention to reduce signal dimensionality and b) still does not make full use of existing knowledge of the signal properties (such as "sparseness" - for example). stICA can be further developed to automatically identify components of relevance - thus dramatically reducing the sub-space dimensions and removing the need for manual intervention, as well as incorporating existing domain knowledge into the source separation process.
The methods will be applied to existing EEG signal databases, with scope for also developing new datasets to specifically test the new techniques being developed.
The University of Warwick provides an inclusive working and learning environment, recognising and respecting every individual’s differences. We welcome applications from individuals who identify with any of the protected characteristics defined by the Equality Act 2010.