GLM General Linear Model

From LONI Pipeline
Jump to: navigation, search

CCB General Linear Model

Overview - General Linear Model

The workflow runs voxelwise statistics on two populations of mice. Each of the populations has been scanned prior to injection of a label and at four subsequent timepoints.

Problem addressed by this workflow

The workflow addresses several problems. First of all, it performs co-registration between the subjects from the two populations and at different timepoints. It then scales all images. Finally, it runs a multiple linear regression module to produce p-values and t-statistics for the two variables, namely cohort and time after injection.

Detailed Workflow Usage & Specifications

Data sources are grouped into the following five module groups: (Wildtype) Pre-Injection Source Wildtype Post-Injection Source (Wildtype) Reference Source Knockout Source Group Sizes (Wildtype + Knockout)

The following briefly describes the protocol:

Wildtype-Preinjection Processing - Using a mouse template as a reference, an MDA is computed from the wildtype pre-injection images; each wildtype pre-injection image is then non-linearly registered to the MDA

Wildtype Processing - The post-injection volumes are registered to the pre-injection volumes, then the transformation matrices generated in the previous step are applied to put all wild-type brains into MDA space

Knockouts Processing - The knockout pre-injection volumes are non-linearly registered to the wildtype pre-injection MDA; then, as in the 'Wildtype Processing' step, the knockout post-injection brains are non-linearly registered to the knockout pre-injection brains and the transformation matrices generated in the previous step are applied to put all knockout brains into MDA space

All co-registered volumes are then scaled, smoothed, and fed into the Multiple Linear Regression module.

Pipeline Workflows

GLM.PNG
  • GLM.pipe: General Linear Model Workflow (01/06/2010)
  • WithinCohort.pipe: Run this after the GLM workflow - with co-registered, smoothed, and scaled volumes for different timepoints within a cohort (01/06/2010)
  • BetweenCohorts.pipe: Run this after the GLM workflow - with co-registered, smoothed, and scaled volumes for the same timepoint from different cohorts (01/06/2010)

Footnotes

  • Outputs and results: The outputs are p-value, t-statistic, b-value, and r-value maps for 'time after injection' and 'cohort'; in particular, we output the raw maps and FDR-corrected p-value maps
  • Expected times: ~2 hours
  • Limitations: Currently, hard-coded to run on 2 groups, each with one pre- timepoint and four post- timepoints
  • Contact person/group: SIG-FLOW Team
  • Tools/packages used in this workflow: FSL, LONI Stats, Minc, AIR