Removal of Scanner Effects in Covariance Improves Multivariate Pattern Analysis in Neuroimaging Data

Chen, Andrew A.; Beer, Jacob; Tustison, Nicholas J.; Cook, Philip A.; Shinohara, Russell T.; Shou, Haochang

doi:10.1101/858415

Cited by 27 publications

(38 citation statements)

References 43 publications

(42 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In brain imaging, more work is needed to identify the factors that contribute to local and distal correlations between vertices, hence inducing a correlation between true associations and "null" vertices, beyond the usual covariates or confounders used in neuroimaging (e.g. MRI scanner/artefact (Chen et al, 2020) or demographics (Montembeault et al, 2012)).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

A parsimonious model for mass-univariate vertex-wise analysis

Couvy‐Duchesne

Zhang

Kemper

et al. 2021

Preprint

View full text Add to dashboard Cite

2.AbstractCovariance between grey-matter measurements can reflect structural or functional brain networks though it has also been shown to be influenced by confounding factors (e.g. age, head size, scanner), which could lead to lower mapping precision (increased size of associated clusters) and create distal false positives associations in mass-univariate vertex-wise analyses. We evaluated this concern by performing state-of-the-art mass-univariate analyses (general linear model, GLM) on traits simulated from real vertex-wise grey matter data (including cortical and subcortical thickness and surface area). We contrasted the results with those from linear mixed models (LMMs), which have been shown to overcome similar issues in omics association studies. We showed that when performed on a large sample (N=8,662, UK Biobank), GLMs yielded large spatial clusters of significant vertices and greatly inflated false positive rate (Family Wise Error Rate: FWER=1, cluster false discovery rate: FDR>0.6). We showed that LMMs resulted in more parsimonious results: smaller clusters and reduced false positive rate (yet FWER>5% after Bonferroni correction) but at a cost of increased computation. In practice, the parsimony of LMMs results from controlling for the joint effect of all vertices, which prevents local and distal redundant associations from reaching significance. Next, we performed mass-univariate association analyses on five real UKB traits (age, sex, BMI, fluid intelligence and smoking status) and LMM yielded fewer and more localised associations. We identified 19 significant clusters displaying small associations with age, sex and BMI, which suggest a complex architecture of at least dozens of associated areas with those phenotypes.

show abstract

Section: Discussionmentioning

confidence: 99%

“…However, brain measurements may exhibit a pattern of correlation, owing to factors (e.g. head size, MRI scanner/artefact (Chen et al, 2020) or demographics (Montembeault et al, 2012)) which can generate confounded brain-trait associations. Induced local correlations with a true brain-biomarker can generate a smear of association (i.e.…”

Section: Introductionmentioning

confidence: 99%

A parsimonious model for mass-univariate vertex-wise analysis

Couvy‐Duchesne

Zhang

Kemper

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The program versions for the methods used below were as follows: [ 13 ], [ 18 ], [ 19 ], [ 20 ], [ 21 ], [ 22 ], [ 23 ], and [ 24 ]. As CovBat was still in development at time of this paper’s publication [ 25 ], its state at the time of this analysis can be replicated by using the GitHub commit , available at (accessed on 9 June 2020).…”

Section: Methodsmentioning

confidence: 99%

Spinal Cord Morphology in Degenerative Cervical Myelopathy Patients; Assessing Key Morphological Characteristics Using Machine Vision Tools

Ost

Jacobs

Evaniew

et al. 2021

JCM

View full text Add to dashboard Cite

Despite Degenerative Cervical Myelopathy (DCM) being the most common form of spinal cord injury, effective methods to evaluate patients for its presence and severity are only starting to appear. Evaluation of patient images, while fast, is often unreliable; the pathology of DCM is complex, and clinicians often have difficulty predicting patient prognosis. Automated tools, such as the Spinal Cord Toolbox (SCT), show promise, but remain in the early stages of development. To evaluate the current state of an SCT automated process, we applied it to MR imaging records from 328 DCM patients, using the modified Japanese Orthopedic Associate scale as a measure of DCM severity. We found that the metrics extracted from these automated methods are insufficient to reliably predict disease severity. Such automated processes showed potential, however, by highlighting trends and barriers which future analyses could, with time, overcome. This, paired with findings from other studies with similar processes, suggests that additional non-imaging metrics could be added to achieve diagnostically relevant predictions. Although modeling techniques such as these are still in their infancy, future models of DCM severity could greatly improve automated clinical diagnosis, communications with patients, and patient outcomes.

show abstract

“…Furthermore, ComBat uses empirical Bayes to learn the model parameters, which assumes that model parameters across features are drawn from the same distribution; this improves the estimation of the parameters where only small sample sizes are available. ComBat has been further developed to include a term explicitly to model for variables of interest to preserve after harmonisation ( Wachinger et al., 2020 ), model covariances ( Chen et al., 2020 ), the incorporation of a generalized additive model (GAM) into the model, extending it to include nonlinear variations ( Pomponio et al., 2019 ), and longitudinal studies ( Beer et al., 2020 ). ComBat, however, is usually applied to the harmonisation of image-derived values and defined associations.…”

Section: Introductionmentioning

confidence: 99%

Deep learning-based unlearning of dataset bias for MRI harmonisation and confound removal

2021

View full text Add to dashboard Cite

Highlights We demonstrate a flexible deep-learning-based harmonisation framework. Applied to age prediction and segmentation tasks in a range of datasets. Scanner information is removed, maintaining performance and improving generalisability. The framework can be used with any feedforward network architecture. It successfully removes additional confounds and works with varied distributions.

show abstract

Removal of Scanner Effects in Covariance Improves Multivariate Pattern Analysis in Neuroimaging Data

Cited by 27 publications

References 43 publications

A parsimonious model for mass-univariate vertex-wise analysis

A parsimonious model for mass-univariate vertex-wise analysis

Spinal Cord Morphology in Degenerative Cervical Myelopathy Patients; Assessing Key Morphological Characteristics Using Machine Vision Tools

Deep learning-based unlearning of dataset bias for MRI harmonisation and confound removal

Contact Info

Product

Resources

About