Fuzzy Clustering fMRI Data

Mark Alexiuk, Institute for Biodiagnostics, National Research Council Canada
(with Nick Pizzi)

Abstract

Magnetic resonance imaging (MRI) is the de facto modality standard for in vivo organ state and structure due to its high resolution images of soft tissues. Functional MRI (fMRI), which infers organ function from blood perfusion, is a series of MRI acquisitions. A model-based approach to fMRI analysis is problematic since the hemo-dynamic response is highly variable. Exploratory data analysis (EDA) techniques such as cluster analysis are often used instead. One such technique is fuzzy C-means (FCM), an iterative alternating optimization algorithm. FCM determines a fuzzy partition (mapping) of the samples to the clusters. Samples belong to each cluster to different degrees. Recently, a novel FCM variant, fuzzy C-means with feature partitions (FCMP) [1], was developed to integrate relationships between features into the clustering process. For instance, one such relationship in fMRI data is the likelihood that spatially proximal voxels have temporally correlated time courses. The utility and efficacy of FCMP is shown in a series of experiments using both synthetic and authentic fMRI data, and benchmarked against EvIdent® [2], a current fMRI research standard, and CHAMELEON [3], a hypergraph partitioning algorithm. Results for noise reduction in the resultant clusters, and discovery of regions of interest are given.

References

[1] M.D. Alexiuk and N.J. Pizzi, "Fuzzy C-means with feature partitions: A spatio-temporal approach to clustering fMRI data," Pattern Recognition Letters, accepted for publication 2005.
[2] N.J. Pizzi, R. Vivanco, and R.L. Somorjai, "EvIdent: A Java-based fMRI data analysis application," Proc. SPIE vol. 3808, pp. 761-770, 1999.
[3] G. Karypis, E.H. Han, and V. Kumar, "CHAMELEON: A Hierarchical clustering algorithm using dynamic modeling," IEEE Computer, vol. 32, no. 8, pp.68-75, August 1999.