The assessment of hemispheric lateralization in functional MRI—Robustness and reproducibility

• Published in: NeuroImage (Orlando, Fla.) Fla.), 2006-10, Vol.33 (1), p.204-217
• Main Authors: Jansen, A., Menke, R., Sommer, J., Förster, A.F., Bruchmann, S., Hempleman, J., Weber, B., Knecht, S.
• Format: Article
• Language: English
• Subjects: Adult; Algorithms; Attention - physiology; Attention deficit hyperactivity disorder; Brain; Brain - anatomy & histology; Brain - physiology; Brain research; Decision Making - physiology; Dominance, Cerebral - physiology; Female; fMRI; Functional Laterality - physiology; Humans; Image Processing, Computer-Assisted; Language; Lateralization; Magnetic Resonance Imaging; Male; NMR; Nuclear magnetic resonance; Psycholinguistics; Reliability; Reproducibility; Reproducibility of Results; Space Perception - physiology; Studies
• ISSN: 1053-8119; 1095-9572
• DOI: 10.1016/j.neuroimage.2006.06.019

Various methods have been proposed to calculate a lateralization index (LI) on the basis of functional magnetic resonance imaging (fMRI) data. Most of them are either based on the extent of the activated brain region (i.e., the number of “active” voxels) or the magnitude of the fMRI signal change. The purpose of the present study was to investigate the characteristics of various variants of these approaches and to identify the one that yields the most robust and reproducible results. Robustness was assessed by evaluating the dependence on arbitrary external parameters, reproducibility was assessed by Pearson’s correlation coefficient. LIs based on active voxels counts at one single fixed statistical threshold as well as LIs based on unthresholded signal intensity changes (i.e., based on all voxels in a region of interest) yielded neither robust nor reproducible laterality results. Instead, the lateralization of a cognitive function was best described by “thresholded” signal intensity changes where the activity measure was based on signal intensity changes in those voxels in a region of interest that exceeded a predefined activation level. However, not all other approaches should be discarded completely since they have their own specific application fields. First, LIs based on active voxel counts in the form of p-value-dependent lateralization plots (LI = LI( p)) can be used as a straightforward measure to describe hemispheric dominance. Second, LIs based on active voxel counts at variable thresholds (standardized by the total number of active voxels) are a good alternative for big regions of interest since LIs based on signal intensity changes are restricted to small ROIs.