Andrew Ryan McIntyre1,2, Xiaowei Song1,3, Evangelos E. Milios2, Malcolm I. Heywood2, Alma Major1, Ryan D'Arcy1,4, Kenneth Rockwood3,5
1Institute for Biodiagnostics - Atlantic, National Research Council, Halifax, NS, Canada; 2Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada; 3Medicinie, Dalhousie University, Halifax, NS, Canada; 4Psychology and Neuroscience, Dalhousie University, Halifax, NS, Canada; 5Centre for Health Care for the Elderly, QEII Health Sciences Centre, Halifax, NS, Canada
A clustering approach is proposed for globally identifying functional connectivity patterns from resting state fMRI data. The Self-Organizing Map algorithm is directly applied to time course data to reduce the input space for efficient application of the k-means algorithm. Hard clusters are selected and correlated time course patterns are assigned to each cluster. Results on two benchmark data sets indicate that the approach is able to efficiently and accurately identify embedded target time course patterns with low variation between initializations.