Skip to content

Latest commit

 

History

History
230 lines (180 loc) · 6.21 KB

draft_course.rst

File metadata and controls

230 lines (180 loc) · 6.21 KB
Raw

Neuroimaging foundation course

Draft schedule

Introduction to neuroimaging ideas with working code.

Developing into analysis stream for full FMRI dataset.

Some introduction to working practice, mainly through example:

  • Version control
  • Testing
  • +- github
  • IPython notebooks (when to use and when not to use)

Reading images Writing images Format issues (nifti, DICOM) Visualization (images and timeseries) DICOM conversion Quality control / diagnostics Slice timing correction (Dealing with bad data points) Realignment with runs Realignment across runs Coregistration within subject across modality (Distortion correction) Spatial normalization (matching between subjects) Diagnostics again (Smoothing) (Adjusting for global scaling) Statistical modeling with the GLM Modeling within run Modeling across runs within subject Modeling across subjects Mixed effects models (combining within and across subjects) Asking questions of a GLM with contrast matrices Inference / multiple comparison correction Family wise error and false discovery rate Permutation / non-parametric inference Region of interest analysis (Machine learning)

Getting ready

Require:

  • python
  • numpy
  • scipy
  • matplotlib
  • ipython notebook (>=0.13.1)
  • sympy
  • nibabel
  • pydicom
  • nipy

Image manipulation

Read a 3D image as a numpy array Show image slices with matplotlib

Read a 4D image as a numpy array Show image slices with matplotlib Get and plot timeseries

Load affine from Nifti file Find 0, 0, 0 (real space) voxel (Visualize image with non-trivial affine) (Show registration of two images with non-trivial affine) (Futz with affine to improve registration)

Plot the timeseries for that voxel using matplotlib

Finding information in the nifti header (e.g. get slice, frequency, phase)

Writing a 3D image Set negative voxels to zero? Set background voxels to zero? Set NaN or Inf voxels to zero? Applying a binary mask to an image? Doing math on images (+ - / *) (covered by mean, std etc below?)

Reading image data from a DICOM image Getting information from the DICOM header DIOCM formats - Siemens, Philips

Basic diagnostics

  • mean image
  • standard deviation image
  • min / max images
  • global signal plots
  • time series differences a la tsdiffana
  • Detecting some common artifacts (RF, slice dropout - with examples)
  • (automated spike detection?)
  • PCA plots / images with nipy?

fMRI preprocessings

Slice timing correction

  • Extract timeseries from first slice and middle slice
  • Plot on same plot
  • Time-series interpolation (NN, linear, spline, sinc) and resampling
  • Some nasty timeseries (spikes) (and what happens with different interpolations)

Realignment

  • On 2D slices
  • The idea of a cost function
  • The idea of 2D resampling
  • Trying different transformations to reduce the cost function (combining cost function and resampling)
  • Finding a best X translation
  • Finding a best X and a best Y translation
  • The idea of the local minimum
  • Extrapolating to 3D
  • Using the best images for realignment * Mean image to time point images for within-run * Mean image to mean image between runs

Distortion correction

  • (the problem at least)

Order of steps

  • Problem of slice timing and realignment
  • Problem of distortion correction relative to both

Spatial normalization

  • Visualizing native space (what is 0, 0, 0)
  • Compare native space between brains
  • Find the central sulcus, calcarine sulcus, Heschl's gyrus) on two brains
  • (Freesurfer surface here? PySurfer?)
  • The idea of MNI space
  • The MNI templates
  • (?The Juelich cytoarchitecture data?)
  • How best to match our two subject's brains?
  • How best to match the brains to the MNI template?
  • (Purpose of normalization)
  • Cost functions for normalization
  • Methods of normalization (shape matching, matching sulci)
  • An affine normalization in nipy (find the best parameters for cost-function, optimization method)
  • (?Lesions and cost-functions?)

Diagnostics

Getting ready for the statistical analysis PCA with nipy again (is there still movement artifact?) (is there still slice-timing artifact?)

Statistics part I: Modelling and testing one voxel / one region

A) First recall simple linear regression

An example with simple regression and independent noise * fitting the model : finding the model's parameters * Noise : properties of the noise - what is a good versus a bad fit ? * Testing one parameter: t-test, F-test

B) First level analyses : case of correlated noise: whitening

  • Going from neural events to BOLD responses -
  • Several experimental conditions: multiple regression
  • Nuisance variable (eg mvt parameters)
  • AR estimation
  • Event related models : one hrf / several basis functions - F-test testing

C) Second level analyses:

  • Why a second level analysis at all ? The idea of several variance components
  • Modelling factors and interactions (between subject factor only), how to interpret the contrasts
  • take the example of a 2 or 3 ways ANOVA model : how do I specify the contrasts ?
  • Mixed effects analyses: within and between factors examples

D) Diagnostic tools ?

  • univariate ?
  • multivariate ?

Statistics part II: multiple comparison issues

  • Family wise error concept: correcting for the number of tests
  • Detection at the voxel level, detection at the cluster level
  • Techniques for FWE correction :
    • Gaussian Random Fields (GRF)
    • Permutation
  • False discovery rate correction: sensitivity and specificity - Effect size, how many subjects, ...
  • Conjunctions ?

Other analyses:

  • machine learning
  • Resting state / activation connectivity (functional, effective ...)
  • ...

Questions

What should we use for volumetric visualization?

  • A crummy matplotlib viewer (we have a few lying around)
  • Eleftherios' fancy new viewer? (need to review) (probably needs Qt / PySide)
  • fslview
  • SPM :)