Design and Analysis for Functional Neuroimaging: A Problem Oriented Approach

 

Many investigators are interested in combining functional, structural, and diffusion MRI data in experiments probing the relationships between brain structure and function. While images from each modality require unique preprocessing techniques, the general principles of experimental design and analysis are similar for all three modalities. This advanced workshop utilizes a unique problem-oriented approach in presenting the principles related to the design and analysis of neuroimaging experiments using structural, functional and diffusion MRI.

 

Organized around a series of case studies, the lectures will cover the theory and practice of task design, data acquisition, preprocessing, statistical modeling, artifact detection, and visualization of data associated with experimental designs of the type routinely used in neuroimaging research. Lectures will be linked with a series of interactive laboratory exercises that will provide participants with experience using SPM8 and MRIcron to combine structural and functional neuroimaging data.

 

Useful course prerequisites include experience with fMRI data analysis, prior attendance at one of our basic SPM courses, and graduate level statistics courses covering multiple regression. Some knowledge of imaging physics will be helpful, but not essential.

 

While the lectures will include in-class exercises using SPM and its extensions, the sessions will also include demonstrations incorporating FSL, Trackvis, FreeSurfer and other software packages. On the final day, we will demonstrate the use of an advanced, Python-based system that allows integration of modules from all of these packages. To be able to participate in the interactive laboratory exercises, students are expected to bring a portable computer with MATLAB, SPM8, MRIcron, and R software already installed.

 

Registration information can be found here.

Information about accomodations for the Boston courses may be found here.

Questions about course arrangements or content should be directed to admin@neurometrika.org

The course instructors are Thomas Zeffiro, Robert Savoy, Susan Whitfield-Gabrieli, Gary Strangman and Satrajit Ghosh.

 

 

TENTATIVE SCHEDULE

 

MONDAY

 

MRI Preprocessing

 

Sources of brain image contrast

  • Structural MRI
  • Functional MRI
  • Diffusion MRI
  • Optimizing tissue contrast in fMRI and diffusion MRI
  • Exercise: measuring image SNR, temporal SNR

 Spatial normalization to a common volume anatomical space

  • Exercise: Direct transformation
  • Exercise: Indirect transformation
  • Exercise: DARTEL transformation

 Voxel-based morphometry preprocessing

  • Exercise: Using SPM8
  • Exercise: Using the VBM8 toolkit

 Surface-based morphometry preprocessing

  • Demonstration: Freesurfer structural preprocessing

LUNCH

Functional MRI preprocessing

  • Geometric distortion correction – susceptibility effects
  • Slice timing correction
  • Realignment Spatial filtering
  • Exercise: Functional MRI preprocessing

 Diffusion MRI preprocessing

  • Geometric distortion correction – eddy currents
  • Realignment
  • Spatial filtering
  • Exercise: DTI preprocessing

 Arterial spin labeled MRI preprocessing

  • Exercise: ASL preprocessing

 

TUESDAY

 

Experimental Design in Neuroimaging

 

Review of the GLM – Regression, ANOVA and ANCOVA

 

Independent measures single factor designs

  • Exercise: Detecting between-group effects on a single task

Dependent (repeated) measures designs

  • Exercise: Detecting within-group effects across multiple tasks

Factorial dependent (repeated) measures designs

  • Exercise: Detecting group by task interactions ANCOVA
  • Exercise: Detecting covariate effects
  • Exercise: Controlling nuisance effects
  • Exercise: Detecting group by covariate interactions

Assumptions underlying ANOVA and ANCOVA

 

LUNCH

 

Background on design efficiency for fixed epoch designs

  • Case study: Epoch design Background on design efficiency for stochastic designs
  • Case study: Stochastic design

Controlling for trial to trial variation in performance

  • Case study: Epoch design with behavioral covariates
  • Case study: Stochastic design with behavioral covariates

Designs to simultaneously detect tonic and phasic effects

  • Case study: Mixed block and stochastic design

Designs to detect parametric task effects

  • Case study: Parametric variation in task properties

Designs to detect group differences in task-related activity

  • Case study: Between-group variable epoch design

Designs to detect group differences in task-related activity controlling for group performance

  • Case study: Between-group stochastic design with performance covariates
  • Case study: Between-group stochastic design with trait covariates

Designs with multiple within-subject repeated measures

  • Case study: Factorial between-group designs

 

WEDNESDAY

 

Voxel-based morphometry and diffusion imaging

 

There is increasing interest in integrating measures of either regional gray matter volume or white matter microstructural properties with measures of regional functional brain activity. In this session we explore techniques for statistical modeling of data derived from structural and diffusion imaging.

 

Introduction to voxel-based morphometry

  • Exercise: Group analysis of gray matter volume

Between group comparisons of brain structure

  • Exercise: Between-group analysis of brain structure

Between group comparisons of brain structure with covariates

 

Controlling for between-group structural differences in fMRI studies

 

Guidelines for reporting a VBM study

 

BREAK

 

Derivation of white-matter microstructure measures

  • Exercise: Derivation of fractional anisotropy, mean diffusivity, radial and axial diffusivity

Assessing between group white-matter-microstructure effects

  • Exercise: Detecting group differences in fractional anisotropy

Assessing within group white-matter microstructure covariate effects

  • Exercise: Detecting covariate effects on fractional anisotropy

Assessing anatomical connectivity with diffusion imaging

  • Exercise: Tractography using structural seeds
  • Exercise: Tractography using functional seeds

 

LUNCH

 

Meta-analysis of functional neuroimaging studies

 

As the functional neuroimaging literature grows, accessing accurate summaries of previous research has become increasingly complex. Quantitative meta-analysis of neuroimaging studies allows objective estimation of experimental effects drawn from existing published datasets. In this session we will describe a range of existing methods for pursuing quantitative meta-analysis of published neuroimage data.

 

Principles of meta-analysis

 

Introduction to ALE meta-analysis

  • Exercise: Meta-analysis of single word reading

Single group meta-analysis

  • Exercise: Meta-analysis of cerebellar regional functional specialization

Between-group meta-analysis

  • Exercise: Meta-analysis of visual processing in autism

Recent developments in neuroimage meta-analysis

 

THURSDAY

 

Design and analysis guidelines for avoiding the consequences of circular analysis

 

Circular analysis can result from the selection of regions for statistical testing based on their response to a prior statistical test, a situation referred to as selection bias. Recently, concerns have been widely raised about the effects of using such "circular" or "non-independent" analyses in functional neuroimaging. We will discuss the strategic and tactical issues related to avoiding problems associated with circular analysis.

 

Statistical significance and effect size

  • Exercise – Computing fMRI effect size estimates

Circular analysis in neuroscience and functional neuroimaging

 

Design strategies for avoiding circular analysis

  • Exercise - Using anatomical regions of interest
  • Exercise – Using functional regions of interest

Analysis tactics for avoiding circular analysis

  • Exercise – Using permutation tests Data exploration and hypothesis testing in neuroimaging

 

LUNCH

 

Using R for behavioral and neuroimaging data analysis

 

Although most fMRI experiments require statistical modeling of the behavioral data collected during image acquisition, few software packages allow modeling of both behavioral and image data. R is a statistical computing language that is rapidly gaining in popularity. Freely available, it is enjoying increased use by neuroimaging investigators and is already supported by a large community of users. In this session we will describe the use of R in modeling both behavioral and imaging data from functional imaging experiments by exploring its mechanisms for data import/export, data modeling and data graphing.

 

R Basics

 

R Commander

 

Basic descriptive statistics using R

  • Exercise – Analyzing response accuracy Regression and ANOVA using R
  • Exercise - Analyzing response time Graphics in R
  • Exercise – Graphing behavioral results from neuroimaging experiments DTI analysis using R fMRI analysis using R

 

FRIDAY

 

Individual and group hemodynamic response modeling

 

While second level (group) statistical parametric maps reveal the spatial patterns of experimental effects, they may not convey the details of these effects. However, it may be desirable to extract and display response properties derived from individual time series. Although SPM offers a variety of methods for visualizing the results of first-level analyses, including beta estimates, adjusted and predicted responses, and event-related responses, these response measures are not readily accessible at the second-level. In this session we will explore a range of issues related to the modeling of hemodynamic response at the first and second level.

 

Individual hemodynamic response modeling

  • Exercise - Using basis set models at the first level
  • Exercise - Using FIR models at the first level

Group hemodynamic response modeling

  • Exercise - Using basis set models at the first-level
  • Exercise - Using FIR models at the first level

Group hemodynamic response analysis using rfxplot

  • Demonstration – Group hemodynamic time course exploration with rfxplot

 

LUNCH

 

Using Python for neuroimaging data analysis

 

The construction of brain image analysis pipelines integrating processing modules from different software packages is difficult using standard scripting languages. Python is an object-oriented computing language that supports rapid, cross-platform development. It is highly scalable and has been extended with a large number of libraries, some of which are specifically designed for neuroimaging applications. After introducing the basics characteristics of the language, we will explore the utility of using Python as a programming environment integrating functions from SPM, Freesurfer and FSL.

 

Python basics

 

RPy – an interface between Python and R

  • Demonstration – fMRI first-level statistical modeling

Nipype - a collaborative platform for neuroimaging software development using a high-level language

  • Demonstration - Creating an analysis pipeline using Nipype
  • Demonstration - Using Nipype to integrate SPM and Freesurfer functions
  • Demonstration - Using Nipype to analyze diffusion data using FSL