Permutation inference for the general linear model

By Anderson M. Winkler, Gerard R. Ridgway, Matthew A. Webster, Stephen M. Smith, Thomas E. Nichols. NeuroImage, 2014

3,477 citations recorded by the source corpus.

Permutation methods can provide exact control of false positives and allow the use of non-standard statistics, making only weak assumptions about the data. With the availability of fast and inexpensive computing, their main limitation would be some lack of flexibility to work with arbitrary experimental designs. In this paper we report on results on approximate permutation methods that are more flexible with respect to the experimental design and nuisance variables, and conduct detailed simulations to identify the best method for settings that are typical for imaging research scenarios. We present a generic framework for permutation inference for complex general linear models (glms) when the errors are exchangeable and/or have a symmetric distribution, and show that, even in the presence of nuisance effects, these permutation inferences are powerful while providing excellent control of false positives in a wide range of common and relevant imaging research scenarios. We also demonstrate how the inference on glm parameters, originally intended for independent data, can be used in certain special but useful cases in which independence is violated. Detailed examples of common neuroimaging applications are provided, as well as a complete algorithm – the “randomise” algorithm – for permutation inference with the glm.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/c83cb156e8812906bf79337032e076f734eb5d80

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Training confounder-free deep learning models for medical applications (2020, methodology): “The HIV effect was tested on each voxel in the GM masks of the c-independent subset by a General Linear Model implemented in Permutation Analysis of Linear Models (PALM, with 5,000 permutations) [10].”
An a contrario approach for the detection of patient-specific brain perfusion abnormalities with arterial spin labelling (2016, neutral): “Using a nonparametric approach, such as permutation-based or sign-flipping technique (Winkler et al., 2014) might be an interesting alternative.”
Faster permutation inference in brain imaging (2016, methodology): “Even though such partitioning is not unique, it can be de ﬁ ned in terms of the contrast C such that the columns of X and Z are orthogonal to each other, and inference on β is equivalent to inference on C ' ψ (Beckmann et al., 2001; Smith et al., 2007; Winkler et al., 2014).”
The impact of traditional neuroimaging methods on the spatial localization of cortical areas (2018, methodology): “uk/fsl/ fslwiki/PALM (32)] can be used without smoothing, or with small amounts of smoothing, in contrast to popular parametric methods, and also work well with HCP-style CIFTI or parcellated data.”
Spontaneous neural encoding of social network position (2017, neutral): “More specifically, FSL’s randomise 51,52 program was used to perform permutation tests and to generate a null distribution of cluster masses for multiple comparisons correction (cluster-forming threshold: p < .01, two-tailed; 5,000 permutations; 10-mm variance smoothing).”
Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates (2015, positive): “To compare the parametric statistical methods used by SPM, FSL, and AFNI to a nonparametric method, all analyses were also performed using a permutation test (22, 23, 27).”
Multimodal Surface Matching with Higher-Order Smoothness Constraints☆ (2017, methodology): “Distributions of LogJ for each subject were then resampled onto the template and compared using FSL’s PALM (Winkler et al., 2014), which performs permutation testing for surface image data.”
Multi-compartment microscopic diffusion imaging (2016, methodology): “Multiple comparison correction across space was carried out by exhaustive permutation testing (Winkler et al., 2014) subject to the masked brain, which then yielded family-wise error (FWE) corrected p-value maps.”
CAT: a computational anatomy toolbox for the analysis of structural MRI data (2024, methodology): “A particu-larl y helpful featur e of the TFCE toolbox is that it automatically r ecognizes exc hangeability bloc ks and potential nuisance par am-eters [ 46 ] from an existing statistical design in SPM.”
Connectivity-based fixel enhancement: Whole-brain statistical analysis of diffusion MRI measures in the presence of crossing fibres (2015, methodology): “Statisticalinferencewas performedusinga general linear model (GLM) and non-parametric permutation testing (Freedman and Lane, 1983; Nichols and Holmes, 2002; Winkler et al., 2014), with 5000 permutations.”
ICA-based artifact removal diminishes scan site differences in multi-center resting-state fMRI (2015, methodology): “FSL (http://www.fmrib.ox.ac.uk/fsl) tools were used for all data analyses (Smith et al., 2004;Woolrich et al., 2009; Jenkinson et al., 2012).”
Robust Permutation Tests For Correlation And Regression Coefficients (2017, methodology): “Winkler et al. (2014) discussed the applications of permutation methods for multiple linear regression to neuroimaging data.”
Investigating white matter fibre density and morphology using fixel-based analysis (2017, methodology): “Family-wise error corrected p-values were assigned to each fixel using non-parametric permutation testing (Holmes et al., 1996; Winkler et al., 2014) with 5000 permutations.”