Hierarchical Geostatistical Analysis in Clustering fMRI Time Series

• Activity Number: 165
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jun Ye*+
• Companies: University of Akron
• Keywords: ill-balanced data ; fMRI time series ; geostatistical clustering

Abstract

The signal of responses in fMRI is typically very small and must be filtered from large amount of noises. As relatively small changes in brain activity are easily buried within noisy measurement, regular clustering algorithms usually do not work well for the ill-balanced fMRI data. Hence cluster analysis has to perform an initial voxel selection procedure at the preliminary stage. I will explore the deficiencies of classical clustering in fMRI and develop a method of hierarchical geostatistical clustering in fMRI time series, effectively combining data reduction and clustering. This method will reorganize unbalanced fMRI data at each iterative step through geostatistical clustering. In addition to enabling robust and accurate analysis for clustering, the proposed new analysis reduces the gap between dimension reduction and clustering and will significantly improve the entire clustering procedure.