Feature Space Analysis for Group Inference in fMRI Data

Recently, fMRI multisubject analysis has emerged as a new topic to embrace the differences in brain response. State-of-the-art multisubject methods - like general linear model (GLM) - suffer from limited sensitivity. A novel method based on feature space fuzzy cluster analysis for fMRI group inference is introduced to overcome this limitation. In the proposed method, a brain tensor is obtained using the cross-correlation analysis of each subject. Then a feature space is constructed from the brain tensor. Fuzzy cluster analysis of the proposed feature space generates a ap of "membership to active cluster”. Statisticalsignificance of the membership map is then assessed using randomization to derive the group inference activation map. This method is applied to experimental and simulated multi-subject fMRI data and results are compared to those of the GLM. We show that the proposed method detects more activated regions in analyzing experimental data and considerably more true positives (30-40%) at all false alarm rates in the simulation study. This means that the proposed method has higher detection sensitivity compared to GLM.

Keywords:

fMRI , group inference , feature space clustering , visual task