Statistical modeling of fNIRS data using the R statistical programming language

Thomas Zeffiro, University of Maryland

Duration: 180 min

Capacity: 30

Level: Advanced, Good coding experience

Statistical Analysis Toolbox

Requirements: Laptop; Matlab, R, and RStudio pre-installed

Synopsis: fNIRS experimental designs are steadily increasing in complexity, often incorporating longitudinal assessments of brain activity. While many fNIRS programs support preprocessing and single-subject statistical modeling, relatively few of them support complex modeling and inference at the group level, particularly when the experimental design includes repeated measures, missing data, covariates and mixed with- and between- subject factors. This short course will begin with an intensive conceptual review of the statistical issues involved in fNIRS single-subject and group analysis. Next, using supplied datasets, we will carry out a series of practical analysis exercises of graded complexity demonstrating the use of R, an open-source statistical programming language, for fNIRS data analysis and visualization. At the completion of the course, participants should be able to begin to use R, RStudio, NeuroconductoR and their associated packages to construct and estimate: 1) single group models, 2) multi-group models, 3) repeated measures models, 4) models including covariates, and 5) models including missing data. The use of longitudinal mixed effects models in designs incorporating both with- and between- subject effects will be covered in detail and a range of data visualization strategies will be demonstrated. Example R code for all the class exercises will be provided, along with a website giving pointers for further independent study.

Rationale: Although fNIRS experimental designs of increasing complexity are becoming more common, many fNIRS analysis packages support only limited group modeling. The R statistical programming environment extends the capabilities of existing preprocessing and analysis packages, providing a rich set of tools for group analysis and results visualization.

Course structure:

  • First level modeling for task designs – example: visuomotor task-related activity
  • Exporting first-level model results from popular fNIRS analysis packages – demonstrations
  • Second level modeling from basic to complex
    • Categorical predictors – example: between group difference in visuomotor task-related activity
    • Continuous predictors – example: behavioral influences on visuomotor task-related activity
    • Longitudinal models – example: change in visuomotor task-related activity following massed practice
  • Designs including within and between factor designs
  • Inference and critical threshold determination – recent controversies and their resolution
  • Visualization
  • Review and questions

Learning objectives:

  • Understand the application of multiple regression to the analysis of single subject fNIRS data.
  • Be able to export first-level image and single-detector results from a range of fNIRS analysis packages, including nirsLAB, HOMER2 and NIRS-SPM.
  • Be able to use the RStudio integrated development environment to construct analysis and visualization workflows.
  • Use R to construct group analysis models to detect single-group effects, between-group effects, covariate effects, and repeated measure effects.
  • Use R to construct longitudinal group analysis models for designs incorporating within and between group effects, and missing data.
  • Use R for model and data visualization.

Requirements: You are expected to bring a laptop with 50 GB of free disk space and a minimum of 8GB RAM, with MATLAB, R, and RStudio installed.