Post-modern fMRI data analysis in parallel imaging
fMRI data analysis relies on a series of complex operations joining several key steps from the image reconstruction to group-level statistical analysis, allowing inference on brain function of healthy or pathological subjects. As far as one step of this pipeline can be improved, it makes sense to measure its impact on the group-level statistical analysis, especially regarding the sensitivity/specificity trade-off. In this context, we have investigated two lines of algorithmic research as well as technological considerations by comparing two multi-channel head coils on the 3T Tim Trio magnet available at NeuroSpin in order to validate whole brain neuroimaging at high spatial resolution while preserving standard temporal resolution. On the one hand, in parallel imaging, we have proposed a wavelet-based regularized and unsupervised SENSE reconstruction pipeline [Châari et al, 2010a,b]. On the other hand, at the within-subject statistical analysis level, we have developed an alternative GLM-based fitting procedure, which has been popularized through the current softwares in the neuroimaging community (SPM,FSL, BrainVISA-fMRI/nipy). Importantly, this alternative embeds an automatic spatially-adaptive regularization process that makes no longer necessary the spatial filtering of the fMRI datasets, which is known to degrade spatial resolution [Vincent et al, 2010; Risser et al, 2010; Ciuciu et al, 2010]. In this talk, we show the impact inherent to each methodological advance on several datasets [Badillo et al, 2010a, Chaâri 2010c] and illustrate situations where choosing the 32 channels coil on the 3T Tim Trio magnet appears more relevant regarding group level statistical results [Badillo et al, 2010b]. Part of these tools will be soon available on the platform through the BrainVISA-fMRI toolbox, on which the advanced JIRFNI training session at Marseille in October, 26-29 will rely.