Pitfalls in the clustering of neuroimage data and improvements by global optimization strategies

In this paper, we examined three vector quantization (VQ) methods used for the unsupervised classification (clustering) of functional magnetic resonan ce imaging (fMRI) data. Classification means that each brain volume element (voxel), according to a given scanning raster, was assigned to one group o f voxels based on similarity of the fMRI signal patterns. It was investigat ed how the VQ methods can isolate a cluster that describes the region invol ved in a particular brain function. As an example, word processing was stim ulated by a word comparison task. VQ analysis methodology was verified usin g simulated fMRI response patterns. It was demonstrated in detail that VQ b ased on global rather than local optimization of the objective function yie lded a higher performance. Performance was measured in statistically releva nt series of VQ attempts using several indices for goodness, reliability an d efficiency of VQ solutions. Furthermore, it was shown that a poor local o ptimization caused either an underestimation or an overestimation of the st imulus-induced brain activation. However, this was not observed if the clus ter analysis was based upon a global optimization strategy. (C) 2001 Academ ic Press.