TY - GEN

T1 - Analyzing brain signals by combinatorial optimization

AU - Dauwels, Justin

AU - Vialatte, François

AU - Weber, Theophane

AU - Cichocki, Andrzej

PY - 2008

Y1 - 2008

N2 - We present a new method to determine the similarity (or synchrony) of a collection of multi-dimensional signals. The signals are first converted into point processes, where each event of a point process corresponds to a burst of activity of the corresponding signal in an appropriate feature space. The similarity of signals is then computed by adaptively aligning the events from the different point processes. If the point processes are similar, clusters containing one point from each time serie will naturally appear. Synchrony is then measured as a function of the size of the clusters and the distance between points within one cluster. The alignment of events is defined in a natural statistical model; the optimal clustering is obtained through maximum a posteriori inference and can be cast as a combinatorial optimization problem. As the dimension and the number of signals increase, so does the complexity of the inference task. In particular, the inference task corresponds to: a) a dynamic program when comparing two 1-dimensional signals b) A maximum weighted matching on a bipartite graph when comparing two d-dimensional signals c) A NP-hard integer program that can be reduced to N-dimensional matching when comparing N ≥ 2 signals We show the applicability of the method by predicting the onset of Mild Cognitive Impairment (MCI) from EEG signals.

AB - We present a new method to determine the similarity (or synchrony) of a collection of multi-dimensional signals. The signals are first converted into point processes, where each event of a point process corresponds to a burst of activity of the corresponding signal in an appropriate feature space. The similarity of signals is then computed by adaptively aligning the events from the different point processes. If the point processes are similar, clusters containing one point from each time serie will naturally appear. Synchrony is then measured as a function of the size of the clusters and the distance between points within one cluster. The alignment of events is defined in a natural statistical model; the optimal clustering is obtained through maximum a posteriori inference and can be cast as a combinatorial optimization problem. As the dimension and the number of signals increase, so does the complexity of the inference task. In particular, the inference task corresponds to: a) a dynamic program when comparing two 1-dimensional signals b) A maximum weighted matching on a bipartite graph when comparing two d-dimensional signals c) A NP-hard integer program that can be reduced to N-dimensional matching when comparing N ≥ 2 signals We show the applicability of the method by predicting the onset of Mild Cognitive Impairment (MCI) from EEG signals.

UR - http://www.scopus.com/inward/record.url?scp=64549098239&partnerID=8YFLogxK

U2 - 10.1109/ALLERTON.2008.4797722

DO - 10.1109/ALLERTON.2008.4797722

M3 - Conference contribution

AN - SCOPUS:64549098239

SN - 9781424429264

T3 - 46th Annual Allerton Conference on Communication, Control, and Computing

SP - 1381

EP - 1388

BT - 46th Annual Allerton Conference on Communication, Control, and Computing

T2 - 46th Annual Allerton Conference on Communication, Control, and Computing

Y2 - 24 September 2008 through 26 September 2008

ER -