Abstract
Conventional analysis of functional magnetic resonance imaging (fMRI) data using the general linear model (GLM) employs a neural model convolved with a canonical hemodynamic response function (HRF) peaking 5 s after stimulation. Incorporation of a further basis function, namely the canonical HRF temporal derivative, accounts for delays in the hemodynamic response to neural activity. A population that may benefit from this flexible approach is children whose hemodynamic response is not yet mature. Here, we examined the effects of using the set based on the canonical HRF plus its temporal derivative on both first- and second-level GLM analyses, through simulations and using developmental data (an fMRI dataset on proprioceptive mapping in children and adults). Simulations of delayed fMRI first-level data emphasized the benefit of carrying forward to the second-level a derivative boost that combines derivative and nonderivative beta estimates. In the experimental data, second-level analysis using a paired t-test showed increased mean amplitude estimate (i.e., increased group contrast mean) in several brain regions related to proprioceptive processing when using the derivative boost compared to using only the nonderivative term. This was true especially in children. However, carrying forward to the second-level the individual derivative boosts had adverse consequences on random-effects analysis that implemented one-sample t-test, yielding increased between-subject variance, thus affecting group-level statistic. Boosted data also presented a lower level of smoothness that had implication for the detection of group average activation. Imposing soft constraints on the derivative boost by limiting the time-to-peak range of the modeled response within a specified range (i.e., 4–6 s) mitigated these issues. These findings support the notion that there are pros and cons to using the informed basis set with developmental data.
Highlights
The most common approach in functional magnetic resonance imaging (fMRI) today is to use a standard general linear model (GLM) regressing the blood oxygen leveldependent (BOLD) signal against predictor variables reflecting expected fluctuations due to the task, for each individual separately, and to report group statistics (e.g., Friston et al, 1995; Worsley and Friston, 1995; Monti, 2011)
We considered the most popular cluster-extent inference (Hayasaka and Nichols, 2003; Woo et al, 2014), which consisted in (i) identifying clusters of contiguous voxels whose intensity exceeded a primary threshold set at the value p < 0.001 in our t-statistic images, and (ii) estimating the probability that any of these clusters occurs as a chance process as derived using distributional approximations from the random field theory (RFT), thereby rejecting those clusters whose probability of being due to chance given the smoothness of our data was highly unlikely—or equivalently thresholding the tstatistic images at p < 0.05 family-wise error (FWE) corrected over all clusters
Simulations of fMRI individual data demonstrated that incorporating the canonical hemodynamic impulse response function (HRF) plus its temporal derivative into the hemodynamic model is meaningless in case only beta estimate related to the canonical HRF would be used at group level, as expected (Calhoun et al, 2004)
Summary
The most common approach in fMRI today is to use a standard general linear model (GLM) regressing the blood oxygen leveldependent (BOLD) signal against predictor variables reflecting expected fluctuations due to the task, for each individual separately, and to report group statistics (e.g., Friston et al, 1995; Worsley and Friston, 1995; Monti, 2011). Several works showed significant variations in the hemodynamic response across brain regions with respect to the overall shape, the time-to-onset and the time-to-peak (Henson et al, 2002; Mohamed et al, 2003; Handwerker et al, 2004; Steffener et al, 2010). Using visuo-motor tasks, Handwerker et al (2004) and Mohamed et al (2003) revealed time-to-onset differences between brain regions, the hemodynamic response being faster in the visual areas and peaking a few milliseconds later in motorrelated areas. BOLD signal magnitude variability is task-dependent, with a larger variability in active tasks compared to more passive ones (Garrett et al, 2013)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
