A POPULATION SEARCH ALGORITHM FOR CLUSTERED MULTIVARIATE SOLUTIONS: APPLICATION TO EEG CONNECTIVITY Ian Daly Institute for Knowledge Discovery Laboratory of Brain-Computer Interfaces Graz University of Technology Krenngasse 37/IV 8010 Graz, Austria ian.daly@tugraz.at Reinhold Scherer Institute for Knowledge Discovery Laboratory of Brain-Computer Interfaces Graz University of Technology Krenngasse 37/IV 8010 Graz, Austria reinhold.scherer@tugraz.at Gernot M ¨ uller-Putz Institute for Knowledge Discovery Laboratory of Brain-Computer Interfaces Graz University of Technology Krenngasse 37/IV 8010 Graz, Austria gernot.mueller@tugraz.at ABSTRACT There is a need to efficiently identify time, frequency and spatial locations between which connectivity occurs within the brain. Therefore, a novel, population based, search al- gorithm is proposed based upon the behaviour of foraging animals. The method is evaluated on a simple grid search prob- lem and on the identification of time-frequency locations of statistically significant phase synchronisation in both syn- thetic and real EEG. The method is shown to be compara- ble to a state-of-the-art phase synchronisation identification algorithm in terms of speed while identifying a large pro- portion of available solutions. KEY WORDS Population search, Connectivity, Electroencephalogram (EEG), Phase locking values (PLV) 1 Introduction Recent years have seen an increasing interest in compu- tational mapping of connectivity patterns in the brain [1]. In particular, it is desirable to identify which brain regions form connections during particular cognitive tasks [2], or during rest [3]. A specific example of this is the phase lock- ing value (PLV) in the electroencephalogram (EEG), which may be used to identify patterns of inter-regional connec- tivity during a range of tasks [4, 5]. Multiple regions may be involved in a number of con- currently occurring connectivity patterns which may be dis- tributed over multiple spatial (cortical) regions, frequency ranges and times, either relative to some event or occur- ring spontaneously. For example, in the case of finger tap- ping, phase synchronisation may be observed in the EEG between the contralateral primary motor cortex (M1) and the supplementary motor area (SMA) at a range of different time and frequency points which may be clustered around a range of distributed time-frequency regions [6]. The traditional method to identify locations - for ex- ample in time, frequency and space - at which connectivity occurs is to simply employ brute force searching and iterate over all possible locations, for example as done in [5, 7]. However, in cases where such structured regions of con- nectivity are thought to exist, but their exact locations are unknown, it is desirable to employ a search method which is efficiently able to identify all regions at which statisti- cally significant connectivity patterns exist. To this end a novel search algorithm is proposed to attempt to identify multiple clustered, distributed regions of connectivity. The algorithm is based upon the metaphor of the search behaviour of a group of foraging animals; modelled by a Lˆ evy flight [8]. The group is initially uniformly dis- tributed across the search space and each animal randomly moves around its immediate area looking for ”food” (lo- cations of statistically significant connectivity). When the location of some connectivity is found the number of ani- mals at that location increases and the location is removed (”eaten”) so that the same location cannot be identified twice. If no significant connectivity is discovered the ani- mals increase the range of their search, spreading out to cover more territory within the search space. This is done by increasing the speed at which the animal moves as the time since they last found the location of some significant connectivity increases. If connectivity locations are still not found by an animal then that animal is removed from the search population. This acts as a stopping criterion on the search. When all the animals have been removed the search ends. The structure of this paper is as follows. Section 2 de- Proceedings of the IASTED International Conference Biomedical Engineering (BioMed 2013) February 13 - 15, 2013 Innsbruck, Austria DOI: 10.2316/P.2013.791-138 175