Glossary

A point of the triangle mesh that FreeSurfer fits to the cortex. Every surface measure has one value per vertex; the fsaverage target has 163,842 vertices per hemisphere.

Also called: node.

The left (lh) or right (rh) half of the cortex. One analysis covers one hemisphere, so a whole-brain study is two analyses.

The surface map analysed at every vertex, such as cortical thickness or surface area. In a formula it carries the qdecr_ prefix: qdecr_thickness.

Also called: vertex variable, surface variable.

The directory that holds one FreeSurfer output directory per subject, named by subject ID, and the target. FreeSurfer calls it SUBJECTS_DIR; QDECR’s argument is dir_subj.

The template subject that every subject’s surface data is resampled to, so that a vertex means the same place for everyone. fsaverage by default.

Also called: template, average subject.

The recon-all -qcache step of FreeSurfer, which resamples each subject’s surface measures to the target and smooths them. QDECR reads the files it writes.

Full width at half maximum: the width, in millimetres, of the smoothing kernel applied to the surface data before the analysis. 10 mm by default, 5 mm for local gyrification.

Also called: smoothing kernel size.

The vertices an analysis covers. By default the cortex of the target, without the medial wall, less any vertex where a subject’s value is exactly zero.

FreeSurfer’s binary format for surface and volume data, with the extension .mgh. QDECR reads subject data from it and writes every result map in it.

A division of the target’s surface into named regions, such as the Desikan-Killiany atlas in aparc.annot. QDECR uses it to say which regions a cluster covers.

Also called: parcellation file.

The same statistical model fitted separately at every vertex, with the vertex measure as its outcome.

Also called: vertexwise, mass-univariate analysis.

The model, in R’s formula notation, with the vertex measure on the left of the tilde: qdecr_thickness ~ age + sex.

The matrix R builds from the formula and the data, with one column per coefficient of the model.

One coefficient of the model, a column of the design matrix, together with every map QDECR writes for it. Stacks are named after their column (age, sexmale) and numbered in the matrix’s order, so stack 1 is usually the intercept.

Also called: contrast.

Observation weights for the regression, one per subject, as in R’s lm.

Several copies of a dataset, each with its missing values filled in differently by multiple imputation. QDECR fits the model to every copy and pools the results.

Also called: imputations.

Combining the estimates from each imputed dataset into one coefficient, standard error and p-value per vertex, with Rubin’s rules.

The correction for multiple testing QDECR applies: vertices that pass a threshold are grouped into clusters, and each cluster is judged by its size against FreeSurfer’s precomputed Monte Carlo simulations of pure noise.

Also called: MCZ correction, Monte Carlo correction.

The vertex-wise p-value a vertex must pass to join a cluster, set with mcz_thr: 0.001 by default.

Also called: MCZ threshold, vertex-wise threshold.

How smooth the residuals of the model are across the surface, estimated after the fit and expressed as a FWHM in millimetres. It picks which of FreeSurfer’s simulations the clusters are compared with.

Also called: estimated FWHM.

A connected set of vertices that all pass the cluster-forming threshold, for one stack.

The probability of a cluster at least as large as the one found, if there were no effect anywhere. A cluster is kept when this is below cwp_thr: 0.025 by default, 0.05 split over two hemispheres.

Also called: CWP.

A cluster whose cluster-wise p-value is below cwp_thr. Only significant clusters appear in the summary and the plots.

The map that numbers each vertex by the significant cluster it belongs to, and is zero elsewhere. FreeSurfer calls it the output cluster number (OCN) map.

Also called: OCN map.

The name of one analysis. QDECR combines it with the hemisphere and the measure, as in lh.age_sex.thickness, for the output directory and file names.

Where an analysis writes its maps, cluster tables and saved result: a directory named after the project, inside dir_out.

The R object an analysis returns, of class vw: the settings, paths, model and summaries. The maps themselves stay in the output directory, and the result knows where.

Also called: vw object, output object.

A matrix stored in a file rather than in memory, which QDECR uses for the vertex data of every subject and for the model’s results while it runs. Built on the bigstatsr package, which calls it an FBM.

Also called: big matrix, FBM.

Where the file-backed matrices are written during an analysis, set with dir_tmp. Fast storage here, such as shared memory, makes the analysis much faster.