Investigation of the effect of different functional MRI sequences on local homogeneity measures : implications for neuroimaging studies

Abstract

Background: Data analysis in functional magnetic resonance imaging (fMRI) is typically performed using model-based methods such as the general linear model, whose detection power is limited by the models’ complexity. The field of fMRI has started adopting data driven approaches. This includes the Vogt-Bailey index, which has not been tested across pulse sequences. Objectives: The study aimed to assess functional activation in fMRI across pulse sequences using the Vogt-Bailey index, and benchmark the results against the general linear model. Methodology: A cohort of 10 research volunteers were scanned in a 3T magnetic resonance imaging (MRI) scanner using 3 separate pulse sequences employing voxels of volumes 1.83mm3 , 23mm3 , and 2.53mm3 respectively while performing a block-design finger tapping experiment. The data was analysed with the Vogt-Bailey index and the general linear model. The brain activation maps obtained for each sequence were compared using the Dice-Sørensen coefficient, and the results from the two techniques were compared using the Overlap coefficient. Results: Applying the Vogt-Bailey index on spatially smoothened data enhanced conformance to the general linear model, with moderately-high Overlap coefficients (0.4 to 0.7). Applying the Vogt-Bailey index on unsmoothened data produced more specific results, exhibiting less conformance to the general linear model, with low-moderate Overlap coefficients (0.2 to 0.3). The functional activation detected by the general linear model was only a subset of that detected with the Vogt-Bailey index. Overall, the results of the general linear model were more reproducible across sequences with moderately-high Dice coefficients (0.6 to 0.7). Conclusions and Recommendations: Spatial smoothing of the 2.53mm3 data may be applied prior to the Vogt-Bailey analysis for higher conformance to the general linear model. Alternatively, for somatotopic evaluations, the Vogt-Bailey analysis may be applied on unsmoothened higher resolution data. The results require further validation across a larger cohort of subjects.