Beyond linearity in neuroimaging: Capturing nonlinear relationships with application to longitudinal studies

Chen, Gang; Nash, Tiffany; Cole, Katherine M.; Kohn, Philip D.; Wei, Shau-Ming; Gregory, Michael D.; Eisenberg, Daniel; Cox, Robert W.; Berman, Karen F.; Kippenhan, J. Shane

doi:10.1016/j.neuroimage.2021.117891

Cited by 9 publications

(9 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As the second derivative of a function is associated with its curvature or concavity, a positive second derivative corresponds to an upwardly concave, while a negative second derivative represents downward concavity. Thus, we control the smoothness or curvature of the HRF h ( t ) by tuning the following penalty term on the integrated square of its second derivative (Wood, 2017; Chen et al, 2021), The metric (4) is considered a natural measure of function smoothness (Wood, 2017), which can be expressed as a regularization process among all the spline weights excluding the first two (corresponding to baseline and linear trend, whose second derivatives are zero): ( β 2 , β 3 , …, β P ) . An example of the impact of this regularization on the original data (Fig.…”

Section: Estimating Population-level Hrf Through Smoothing Splinesmentioning

confidence: 99%

“…First, as the first two splines among the thin plates are overall mean and linearity (Fig. 2B), the penalty (4) on β is essentially an L 2 regularization imposed on those spline weights other than the first two elements, β 0 and β 1 (Wood, 2017; Chen et al, 2021). That is, the information is regularized across the nonlinear elements ( β 2 , β 3 , …, β P ) ′ .…”

Section: Estimating Population-level Hrf Through Smoothing Splinesmentioning

confidence: 99%

See 1 more Smart Citation

BOLD response is more than just magnitude: improving detection sensitivity through capturing hemodynamic profiles

Chen

Taylor

Reynolds

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Typical FMRI analyses assume a canonical hemodynamic response function (HRF) with a focus on the overshoot peak height, while other morphological aspects are largely ignored. Thus, in most reported analyses, the overall effect is reduced from a curve to a single scalar. Here, we adopt a data-driven approach to HRF estimation at the whole-brain voxel level, without assuming a profile at the individual level. Then, we estimate the BOLD response in its entirety with a smoothness constraint at the population level to improve predictive accuracy and inferential efficiency. Instead of using just the scalar that represents the effect magnitude, we assess the whole HRF shape, which reveals additional information that may prove relevant for many aspects of a study, as well as for cross-study reproducibility. Through a fast event-related FMRI dataset, we demonstrate the extent of under-fitting and information loss that occurs when adopting the canonical approach. We also address the following questions: 1) How much does the HRF shape vary across regions, conditions, and clinical groups? 2) Does an agnostic approach improve sensitivity to detect an effect compared to an assumed HRF? 3) Can examining HRF shape help validate the presence of an effect complementing statistical evidence? 4) Could the HRF shape provide evidence for whole-brain BOLD response during a simple task?

show abstract

Section: Estimating Population-level Hrf Through Smoothing Splinesmentioning

confidence: 99%

Section: Estimating Population-level Hrf Through Smoothing Splinesmentioning

confidence: 99%

BOLD response is more than just magnitude: improving detection sensitivity through capturing hemodynamic profiles

Chen

Taylor

Reynolds

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Therefore, domain adaptation should be well-suited to MRI data harmonization by creating features that are indiscriminate with respect to the scanner, but correctly discriminate with respect to the data features of interest such as case-control status, age, etc. (Chen et al, 2020b). Convolutional neural networks (CNN), which are popular and well-adapted to vision problems, have also been deployed for data harmonization with demonstrated success at age prediction, although CNN performance can be susceptible to registration related artifacts (Dinsdale et al, 2021).…”

Section: Discussionmentioning

confidence: 99%

A Comparison of Methods to Harmonize Cortical Thickness Measurements Across Scanners and Sites

Sun

Rakesh

Haswell

et al. 2021

Preprint

View full text Add to dashboard Cite

Results of neuroimaging datasets aggregated from multiple sites may be biased by site-specific profiles in participants demographic and clinical characteristics, as well as MRI acquisition protocols and scanning platforms. We compared the impact of four different harmonization methods on results obtained from analyses of cortical thickness data: (1) linear mixed-effects model (LME) that models site-specific random intercepts (LMEINT), (2) LME that models both site-specific random intercepts and age-related random slopes (LMEINT+SLP), (3) ComBat, and (4) ComBat with a generalized additive model (ComBat-GAM). Our test case for comparing harmonization methods was cortical thickness data aggregated from 29 sites, which included 1,343 cases with posttraumatic stress disorder (PTSD) (6.2-81.8 years old) and 2,067 trauma-exposed controls without PTSD (6.3-85.2 years old). We found that, compared to the other data harmonization methods, data processed with ComBat-GAM were more sensitive to the detection of significant case-control differences in regional cortical thickness (Chi2(3) = 34.339, p < 0.001), and case-control differences in age-related cortical thinning (Chi2(3) = 15.128, p = 0.002). Specifically, ComBat-GAM led to larger effect size estimates of cortical thickness reductions (corrected p-values < 0.001), smaller age-appropriate declines (corrected p-values < 0.001), and lower female to male contrast (corrected p-values < 0.001) in cases compared to controls relative to other harmonization methods. Harmonization with ComBat-GAM also led to greater estimates of age-related declines in cortical thickness (corrected p-values < 0.001) in both cases and controls compared to other harmonization methods. Our results support the use of ComBat-GAM for harmonizing cortical thickness data aggregated from multiple sites and scanners to minimize confounds and increase statistical power.

show abstract

“…6b ). Specifically, we adopted thin plate splines as basis functions in a multilevel model to adaptively accommodate the nonlinearity of each cross-layer profile 107 . The measurement uncertainty (standard error) of the VASO response was incorporated as part of the input in the model, which was numerically solved through the R package mgcv 108 to obtain the estimated cross-layer VASO profiles and their uncertainty bands.…”

Section: Methodsmentioning

confidence: 99%

Layer-specific, retinotopically-diffuse modulation in human visual cortex in response to viewing emotionally expressive faces

Chai

et al. 2022

Nat Commun

View full text Add to dashboard Cite

Viewing faces that are perceived as emotionally expressive evokes enhanced neural responses in multiple brain regions, a phenomenon thought to depend critically on the amygdala. This emotion-related modulation is evident even in primary visual cortex (V1), providing a potential neural substrate by which emotionally salient stimuli can affect perception. How does emotional valence information, computed in the amygdala, reach V1? Here we use high-resolution functional MRI to investigate the layer profile and retinotopic distribution of neural activity specific to emotional facial expressions. Across three experiments, human participants viewed centrally presented face stimuli varying in emotional expression and performed a gender judgment task. We found that facial valence sensitivity was evident only in superficial cortical layers and was not restricted to the retinotopic location of the stimuli, consistent with diffuse feedback-like projections from the amygdala. Together, our results provide a feedback mechanism by which the amygdala directly modulates activity at the earliest stage of visual processing.

show abstract

Beyond linearity in neuroimaging: Capturing nonlinear relationships with application to longitudinal studies

Cited by 9 publications

References 24 publications

BOLD response is more than just magnitude: improving detection sensitivity through capturing hemodynamic profiles

BOLD response is more than just magnitude: improving detection sensitivity through capturing hemodynamic profiles

A Comparison of Methods to Harmonize Cortical Thickness Measurements Across Scanners and Sites

Layer-specific, retinotopically-diffuse modulation in human visual cortex in response to viewing emotionally expressive faces

Contact Info

Product

Resources

About