Treatment of ill-balanced datasets of fMRI with Modified Fuzzy c-means Method

In fMRI dataset, the population of actived voxels is always much less than the total population of the voxels, and that produced an ill-balanced dataset. Some methods, such as limiting the analysis to the gray matter voxels where the BOLD signal is expected and removing the voxels that is absolutely non-actived based on statistical criteria, have been used to treat the ill-balanced dataset. In this article, a new method, modified fuzzy c-means (MFc), has been proposed to treat the ill-balanced dataset of fMRI. The main difference from other statistical methods is that it is data-driven. The MFc method is used to classify the voxels into two clusters with nearly the same population and all actived voxels are contained in one cluster. Thus we got nearly half voxels to analysis and the ill-balanced dataset can be treated. The efficiency of clustering analysis is also boosted