Batch Programming of fMRI Data Analysis

1
Batch Programming of fMRI Data Analysis

• Lars Kasper Christoph Mathys

2
Overview

• Introduction Example Dataset
• General fMRI Data Analysis Workflow with SPM
• Quality Assessment of Raw Data
• Spatial Preprocessing
• Statistical Design The General Linear Model
• Results Analyzing Contrast Reporting
• Within-Subject Batching (Single Subject)
• Subject-independent Analysis Steps
• Subject-independent Data Flow (Dependencies)
• Subject-related data
• Between-Subject-Batching (Multiple Subject)

3
Overview

• Introduction Example Dataset
• General fMRI Data Analysis Workflow with SPM
• Quality Assessment of Raw Data
• Spatial Preprocessing
• Statistical Design The General Linear Model
• Results Analyzing Contrast Reporting
• Within-Subject Batching (Single Subject)
• Subject-independent Analysis Steps
• Subject-independent Data Flow (Dependencies)
• Subject-related data
• Between-Subject-Batching (Multiple Subject)

4
Overview of SPM
Statistical parametric map (SPM)
Image time-series
Design matrix
Kernel
Realignment
Smoothing
General linear model
Gaussian field theory
Statistical inference
Normalisation
p lt0.05
Template
Parameter estimates
5
What is batch processing?

• Repeats same data analysis for many subjects
  (gt2)
• Not prone to human errors, reproducible what was
  done
• e. g. jobs mat-files
• Runs automatically, no supervision needed
• Researcher can concentrate on assessing the
  results
• CAVEAT Tempting to forget about all analysis
  steps in between which could lead to errors in
  your conclusions
• Therefore Always make sure, that meaningful
  results were created at each step
• Using Display/CheckReg to view raw data,
  preprocessed data
• Using spm_print to save reported supplementary
  data output
• If anything went wrong, use debugging

6
3 flavors of batching Goals of this tutorial

• After finishing this session, you will be able to
  
• analyze fMRI datasets using
• the Graphical User Interface (GUI) of SPM
• The Batch Editor of SPM
• A template Matlab .m-script file to batch very
  flexibly

2
1
3
7
Introducing the Dataset

• Rik Hensons famous vs non-famous faces dataset
• http//www.fil.ion.ucl.ac.uk/spm/data/face_rep/fa
  ce_rep_SPM5.html
• Includes a manual with step-by-step instruction
  for analysis (homework -))
• Download from SPM homepage (available for SPM5,
  but works fine with SPM8)

8
Introducing the Dataset

• Factorial 2 x 2 design to investigate repetition
  suppression
• Question Influence of repeated stimulus
  presentation on brain activity (accomodation of
  response)?
• Each stimulus (pictures of faces) presented twice
  during a session
• Condition Rep, Level 1 or 2
• lag between presentations randomized
• 26 Famous and 26 non-famous faces to
  differentiate between familiarity (long-term
  memory) and repetition
• Condition Fam, Level F(amous) and N(onfamous)
• Task Decision whether famous or nonfamous
  (button-press)

9
Introducing the Dataset Published Results

• Right Fusiform face area
• Repetition suppression for familiar/famous faces
• Left Occipital face area (posterior, occip.
  extrastriate)
• Repetition suppression for familiar AND
  unfamiliar faces
• Posterior cingulate and bilateral parietal cortex
• Repetition enhancement

10
Overview

• Introduction Example Dataset
• General fMRI Data Analysis Workflow with SPM
• Quality Assessment of Raw Data
• Spatial Preprocessing
• Statistical Design The General Linear Model
• Results Analyzing Contrast Reporting
• Within-Subject Batching (Single Subject)
• Subject-independent Analysis Steps
• Subject-independent Data Flow (Dependencies)
• Subject-related data
• Between-Subject-Batching (Multiple Subject)

11
Spatial Preprocessing Realign
Batch Editor
Batch File
GUI

• sd

FORMAT P spm_realign (P,flags)
12
Spatial Preprocessing Unwarp
Batch Editor
Batch File
GUI
uw_params spm_uw_estimate (P,uw_est_flags) spm_
uw_apply (uw_params,uw_write_flags)
13
Uhthis takes ages

• Now you can probably value the benefits of batch
  processing. If you are still keen on doing all
  that by hand (good exercise!), refer to the
  following
• The SPM manual
• Most current version in your spm8-folder,
  sub-folder man/manual.pdf
• Rik Hensons famous vs non-famous faces dataset
• http//www.fil.ion.ucl.ac.uk/spm/data/face_rep/fa
  ce_rep_SPM5.html
• Included in SPM manual, chapter 29, with
  step-by-step instruction for analysis
• Available for SPM5, but works fine with SPM8

14
Overview

• Introduction Example Dataset
• General fMRI Data Analysis Workflow with SPM
• Quality Assessment of Raw Data
• Spatial Preprocessing
• Statistical Design The General Linear Model
• Results Analyzing Contrast Reporting
• Within-Subject Batching (Single Subject)
• Subject-independent Analysis Steps
• Subject-independent Data Flow (Dependencies)
• Subject-related data
• Between-Subject-Batching (Multiple Subject)

15
General Workflow for the batch interface
3
1

• Top-down approach
• Specify subject-independent data/analysis steps
• Specify subject-independent file-dependencies
  (data flow)
• Specify subject-related data (e.g. event-timing)

1
2
2
3
16
1. The subject-independent analysis parts

• Load all modules first (in right order!)
• Then specify details (where Xs are found) which
  are subject independent
• TR
• Nslices
• model factors
• contrasts of interest

17
2. Data-flow specification (subject-independent
dependencies)

• Specify, which results of which steps are input
  to another step (DEP-sign)
• e.g. smoothed images needed for model spec
• Afterwards save this job as template .mat file

18
3. Add subject-dependent data/information

• Essentially go to all Xs and fill in appropriate
  values
• e.g. the .mat-file of the conditions
  onsets/durations
• Save this job as subject-batch file Run

19
Overview

• Introduction Example Dataset
• General fMRI Data Analysis Workflow with SPM
• Quality Assessment of Raw Data
• Spatial Preprocessing
• Statistical Design The General Linear Model
• Results Analyzing Contrast Reporting
• Within-Subject Batching (Single Subject)
• Subject-independent Analysis Steps
• Subject-independent Data Flow (Dependencies)
• Subject-related data
• Between-Subject-Batching (Multiple Subject)

20
Between-Subject-Batching (Multiple Subject)
1
3
2
1

• Make sure, parameters to be adjusted have an X
  (clear value) for the single subject template
• Specify a meta-job with Run batch
• Create one run for every subject and add missing
  parameter values (in right order)

2
3
21
Resources and Useful Literature

• All step-by-step instructions can be found in the
  SPM manual, chapter 40
• Also multiple-session and multiple subjects
  processing included
• The SPM helpline/mailing list
• E.g. bug precluding the batch-file selector form
  working was fixed here, but not in the updates
  yet https//www.jiscmail.ac.uk/cgi-bin/webadmin?A2
  ind1001LSPMPR39357
• Batch templates are in your spm path
• Configured subject-independent analysis steps
• ltspm8gt/man/batch/face_single_subject_template_nod
  eps.m
• With dependencies included
• ltspm8gt/man/batch/face_single_subject_template.m
  
• With multiple subjects
• ltspm8gt/man/batch/face_multi_subject_template.m

22
Many, many thanks to

• Klaas Enno Stephan
• The SPM developers (FIL methods group)

23
Extending the batchfile with SPM GUI functions

• Debugging
• Generally a good idea to find out how things work
  in SPM
• Crucial for batch-programming using a .m-file
• Here debug spm.m by setting a breakpoint
• If called function found, use edit
  ltfunctionnamegt.m to look at the comments in the
  file

24
Tuning the engine Matlab workspace variables

• e.g. to manipulate SPM.mat or jobs by hand
• also important during debugging, how variables
  are defined and changed