Robust clustering of epileptic spikes using graph-theoretic algorithms

Considers the problem of clustering multichannel EEG epileptic spikes. The spikes are assumed to belong to one of a number of clusters or to be outliers. The presence of outliers calls for robust algorithms, which is why hierarchic graph-theoretic techniques have been used. As a result they give a criterion function, measuring the distinctness of the clusters, as a function of the hierarchy level and the graph parameter value k. The problem is to determine the most suitable level and k value, which should be chosen such that the criterion function have an abrupt decrease for the next lower value. The authors propose that the decision is made using the criterion function surface as an interface to a human