MARGM: A multi-subjects adaptive region growing method for group fMRI data analysis

Abstract Region growing has been utilized in the analysis of functional magnetic resonance imaging (fMRI) data for many years, while some influential factors, such as the definition of growing criterion and the selection of threshold have a great impact on the application and development of region growing. In this paper, a multi-subjects adaptive region growing method (MARGM) is proposed for the group fMRI analysis, where initial seed-region of multi-subjects is automatically determined by combining the split-merge based seed-region selection method with a prior template. Next, an improved homogeneity criterion is defined to control region growing, and an adaptive threshold is introduced to accelerate the convergence process of region growing. Finally, reproducible coefficients are used to obtain spatial information at the group level by considering temporal information simultaneously. The experimental results of hybrid-data and real fMRI data showed that MARGM could generate effective and reliable results compared to the classical fMRI data analysis methods including group independent component analysis and statistical parametric mapping. It could be used to obtain more representative information at the group level, which reflected the commonalities of the subjects in the group. Therefore, it is expected to have wide applicability for the fMRI data analysis.