Introduction

Overview of the workshop

After an introductory presentation, it starts with basic operations of reading, writing, selecting, and aggregating dataset structures. This is followed by MVPA correlation and classficiation analysis of fMRI data in a region of interest. Subsequently, this is extended to exploratory searchlight analysis, representational similarity analysis, MEEG analysis in the space and time dimensions, and surface-based searchlights . Finally approaches to multiple comparison are discussed.

Format

In this workshop, all material is present on the website. Each exercise part of the workshop has three parts:

• short presentation and introduction to exercise

• time to work on the exercise

• presentation of a possible solution to the exercise

Exercises are provided in the form of code skeletons, with part of the code left out as an exercise. Full solutions for all exercises are provided on the website.

Prerequisites

• Matlab / Octave advanced beginner level. experience.

• fMRI and/or MEEG advanced beginner level analysis experience.

• Working Matlab or Octave installation.

• Working FieldTrip installation (required for MEEG analysis).

• MRI data viewer, such as MRIcron (strongly recommended).

• CoSMoMVPA source code and tutorial data.

• It is recommended, prior to the course, to:

  • read the CoSMoMVPA manuscript (doi:10.1101/047118, citation [OCH16]).

  • have the most recent CoSMoMVPA code (see Download instructions).

  • have a recent version of the tutorial data.

  • have set paths properly in .cosmomvpa.cfg (described here)

  • have tested that you can load and save data from and to the paths in .cosmomvpa.cfg.

Goals of this course

• Learn typical MVPA approaches (correlation analysis, classification analysis, representational similarity analysis).

• Learn how these approaches can be applied to both fMRI and MEEG data.

• Learn how to use CoSMoMVPA to perform these analyses:

  • Understand the dataset structure to represent both the data itself (e.g. raw measurements or summary statistics) and its attributes (e.g. labels of conditions (targets), data acquisition run (chunks).

  • See how parts of the data can be selected using slicing and splitting, and combined using stacking.

  • Introduce measures that compute summaries of the data (such as correlation differences, classification accuracies, similarity to an a prior defined representational simillarity matrix) that can be applied to both a single ROI or in a searchlight.

• Learn multiple-comparison approaches.

• Make yourself an independent user, so that you can apply the techniques learnt here to your own datasets.

Not covered in this course

• Preprocessing of fMRI / MEEG data

• Learning to use Matlab / Octave

• Dataset types other than volumetric and surface-based fMRI data and MEEG time-locked data. (Not covered: source-space MEEG)

• How to become a CoSMoMVPA developer

Datasets

For most of the course we will be using the AK6 dataset and the MEG obj6 dataset (described below). Although these can be downloaded separately, it is recommended however to use the full tutorial dataset.

Download link: full tutorial data.

AK6 dataset

This dataset is used for exercises shown on the website (with answers), and you can use it to learn MVPA. It contains preprocessed data for 8 subjects from [CGG+12]. In this experiment, participants were presented with categories of six animals: 2 primates: monkeys and lemurs; 2 birds: mallard ducks and yellow-throated warblers; and 2 bugs: ladybugs and luna moths.

Download link: tutorial data with AK6 data only

For each participant, the following data is present in the ak6 (for Animal Kingdom, 6 species) directory:

- s0[1-8]/                  This directory contains fMRI data from 8 of the 12
                            participants studied in the experiment reported in
                            Connolly et al. 2012 (Code-named 'AK6' for animal
                            kingdom, 6-species). Each subject's subdirectory
                            contains the following data:
   - glm_T_stats_perrun.nii A 60-volume file of EPI-data preprocessed using
                            AFNI up to and including fitting a general linear
                            model using 3dDeconvolve. Each volume contains the
                            t-statistics for the estimated response to a one
                            of the 6 stimulus categories. These estimates were
                            calculated independently for each of the 10 runs
                            in the experiment.
   - glm_T_stats_even.nii   Data derived from glm_T_stats_perrun.nii.
   - glm_T_stats_odd.nii    Each is a 6-volume file with the T-values averaged
                            across even and odd runs for each category.
   - brain.nii              Skull-stripped T1-weighted anatomical brain image.
   - brain_mask.nii         Whole-brain mask in EPI-space/resolution.
   - vt_mask.nii            Bilateral ventral temporal cortex mask similar to
                            that used in Connolly et al. 2012.
   - ev_mask.nii            Bilateral early visual cortex mask.

Also present are model similarity structures, which you can see here:

This data is stored in the models directory:

- models
   - behav_sim.mat          Matlab file with behavioural similarity ratings.
   - v1_model.mat           Matlab file with similarity values based on
                            low-level visual properties of the stimuli.

MEG obj6 dataset

This dataset is used for both the tutorial and for the assignments.

It contains MEG data from a single participant viewing images of six categories; for details see the README file.

Download link: tutorial data with MEEG obj6 data only.

Tentative schedule

When:

• 22.07.2019 - 26.07.2019.

Where:

• Magdeburg, Germany; Universitätsplatz campus, Gebäude 28, room 27

For dinner and other information, see: https://www.noesseltlab.org/events-presentations/3rd-modelling-symposium-1/

Date and time	Description
Monday
09:00	General introduction presentation
10:30	Coffee break
11:00	Getting started. Get Started; Dataset basics
12:30	Lunch break
14:00	Dataset basics
15:30	Coffee break
16:00-17:30	Split-half correlations. Split-half correlation-based MVPA with group analysis
17:40-18:30	Optional: discuss your data models
Tuesday
09:00	Classification analysis, Classification analysis with cross-validation.
10:30	Coffee break
11:00	Using CoSMoMVPA measures
12:30	Lunch break
14:00	Double dipping, Using CoSMoMVPA neighborhoods for regions of interest first part
15:30	Coffee break
16:00-17:30	Neighborhoods and searchlight basics. Using CoSMoMVPA neighborhoods for regions of interest
17:40-18:30	Optional: discuss your data models
Wednesday	Free day
Thursday
09:00	Whole-brain fMRI searchlight. Use the searchlight with a neighborhood and a measure
10:30	Coffee break
11:00	M/EEG searchlight part 1: General MEEG analysis toolboxes
12:30	Lunch break
14:00	M/EEG searchlight part 2: General MEEG analysis toolboxes
15:30	Coffee break
16:00-17:30	M/EEG time generalization: MEEG time generalization
17:40-18:30	Optional: discuss your data models
Friday
09:00	Present your data
10:30	Coffee break
11:00	Representational similarity analysis Representational similarity analysis
12:30	Lunch break
14:00	Surface-based searchlight. Surface-based fMRI searchlight
15:30	Coffee break
16:00-17:30	Multiple comparison correction. Concluding remarks. Using CoSMoMVPA multiple-comparison correction

Acknowledgements

Thanks to Felix Ball, Emanuele Porcu, Peter Vavra, Nico Marek, Camila Agostino, Tömme Noesselt for organizing the symposium.

Contact

Please send an email to a@b, b=gmail.com, a=n.n.oosterhof.

Back to index