Adaptive experimental condition selection in event-related fMRI

Abstract

International audienceStandard Bayesian analysis of event-related functional Magnetic Resonance Imaging (fMRI) data usually assumes that all delivered stimuli possibly generate a BOLD response everywhere in the brain although activation is likely to be induced by only some of them in specific brain areas. Criteria are not always available to select the relevant conditions or stimulus types (e.g. visual, auditory, etc.) prior to estimation and the unnecessary inclusion of the corresponding events may degrade the results. To face this issue, we propose within a Joint Detection Estimation (JDE) framework, a procedure that automatically selects the conditions according to the brain activity they elicit. It follows an improved activation detection that we illustrate on real data

Group-level impacts of within- and between-subject hemodynamic variability in fMRI

“…As a consequence, small temporal differences in condition-specific TTP or FWHM might reflect a differential neural activity either due to the adaptive tuning of neuronal efficacy or to the recruitment of different neuronal populations. The JDE should be extended to answer this issue and for doing so one future direction might consist of first detecting parcel-specific relevant conditions as proposed in [3] and then associating each relevant condition with a specific HRF shape.…”

Variational variable selection to assess experimental condition relevance in event-related fMRI

“…Our use of binary variables is then rather oriented toward the selection of stimulus types which has to be done across the whole set of voxelwise regressions (Section 2). The proposed approach is carried out in a variational EM framework (Section 3), which offers a faster alternative to intensive stochastic procedures as used in [7]. Simulated experiments confirm the ability of our model to select the relevant conditions while real fMRI data illustrate an enhanced determination of activated brain regions (Section 4).…”

“…Our parsimonious model improves both detection and estimation compared to the complete model. An advantage over the selection made in [7] is an easier choice for sigmoid parameters τ 1 and τ 2 that can be set in practice independently of the parcel size and the number of activated voxels, allowing in particular to detect small clusters with a high activation. Eventually, further real data analysis would be necessary for an extended study with a particular emphasis on the group-level impact of parcel-wise adaptive definition of parsimonious models [11].…”

“…In this case, we consider that such evoked activity is artifactual and decide that the stimulus type should be discarded from the model definition. This is different from [7] in which a condition is considered as irrelevant when the number of activated voxels is too small. As confirmed by the experiments below, defining an appropriate relevance threshold appears to be easier and less data dependent for the average activation amplitude criterion.…”