Neurohacking is the continuous process of using, improving and designing the simplest open source scripted software that depends on the minimum number of software platforms and is dedicated to improving the correctness, reproducibility, and speed of neuroimage data analysis.

Theme 1: Overview and Set Up

  • Structural MRI:
    • High spatial resolution
    • Reveals the anatomic structure of soft tissues
    • Used extensively in clinical and research practice
    • Versatile: different contrasts can target different tissue types : FLAIR, T1, T2
    • Sensitive to pathology (e.g. brain cancer, Multiple Sclerosis lesions)

Theme 2: Data Structure and Operations

2.1 DICOM

DICOM format, two components:

  • image data (in pixels, matrix)
  • header (meta-data)

Note: One DICOM file = one “slice” of the brain.

Use the function readDICOM in oro.dicom package in R.

summary: Image data are stored as collection of 2D slice files for DICOM format.

2.2 NIfTI

NifTI:

  • standardized representation of images
  • 3D array: stacking individual slices on top of each other.

DICOM vs NIfTI

  DICOM NifTI
File extenstion .dcm .nii.gz
File representation one slice of the brain 3D image of the brain
Header contains Many information image meta-data, no patient info
Storage Different folders per subject, complex data structures Different files can be in the same directory

Some useful functions to process NifTI format:

  • Use the function dicom2nifti in the oro.dicom package to convert DICOM format to NIfTI format.
  • Use the writeNifTI and readNIfTI functions in the oro.nifti package to write and read NIfTI files.

2.3 Visualization

Use the following functions from oro.nifti packages to visualize the brain images.

  • image: display NifTI data
  • orthographic: display 3-planes of an image.
  • overlay: display overlay of two images, NAs are not plotted. (We typically create a mask image before the overlay)

2.4 Data Manipulation

  • masking
  • basic operations (addition, subtraction, multiplication)

2.5 Transformation and Smoothing

Transformation

  • Linear/ spline transfer functions
  • Used for better visualization, prediction and input into a standard software.

Smoothing

  • Use AnalyzeFMRI::GaussSmoothArray to apply a Gaussian smooth to the image.

2.6 Basic MRI contrasts

  • T1
  • T2
  • FLAIR

Theme 3: Preprocessing

Overview:

  • Pre-processing pipelines
  • Inhomogeneity correction
  • Skull stripping
  • Registration: rigid/affine/nonlinear
  • Intensity normalization

3.1 Preprocessing

Pipeline

DICOM -> NIfTI -> N3 Correction -> Skull Strip -> Registration

Inhomogeneity

The probability distribution function of tissue class intensities should not depend on the spatial localization of the tissue.

Skull-stripping

Remove extra-cerebral voxels from the volume.

3.2 Image Registration

Registration

A spatial transformation of images with the goal of making locations (voxels, ROIs) have the same or similar interpretation.

Co- Registration

Co-register volumes from different modalities to one another.

  • Cross-sectional between-modalities
  • Longitudunal within-modality
  • Longitudinal between-modalities

Registration to a template

  • MNI template
  • JHU_Eve template

Complexity

  • rigid (6df)
  • affine (12df)
  • nonlinear (>12df)

Rigid

  • T_rigid (v) = Rv + t
  • R is rotation matrix, and t is translation vector.

Affine

  • T_affine(v) = Av + t
  • A is not constrained to be a rotation matrix.

3.3 fslr

  • Virtual machine
  • fslr package

3.4 Inhomogeneity correction using fslr

  • Use function fslr:: fsl_biascorrect 
    fast_img = fsl_biascorrect(nim, retimg = TRUE)

3.5 Brain extraction using fslr

  • Use function fslr:: fslbet 
    bet_fast = fslbet(infile = fast_img, retimg= TRUE)

3.6 Image registration using fslr

  • From FSL: FLIRT(FMRIB’s Linear Image Registration Tool) is an automated and robust tool for linear(rigid, affine) brian image registration.
  • Use function fslr:: flirt
## rigid
registered_fast = flirt(infile= bet_fast2, reffile = template, dof = 6, retimg = TRUE)

## affine
reg_fast_affine= flirt(infile= bet_fast2, reffile = template, dof = 12, retimg = TRUE)

Use function fslr::fnirt_with_affine to perform nonlinear image registration and affine registration before that.

fnirt_fast= fnirt_with_affine(infile= bet_fast2, reffile = template, outfile= "FNIRT_to_Template", retimg = TRUE)

3.7 ANTsR

  • AnTsR: port of ANTS into R using Rcpp
  • focus on image Inhomogeneity correction (N3 and N4), image registration

3.8 Basic Data Manipulation with ANTsR

  • reading images: antsImageRead. Note: the extension of the filename (.nii.gz) must be specified; the dimension of the image must be supplied (2, 3, or 4)

  • ANTsR images: the antsImage class
  • Basic Statistics: mean, as.array etc
  • ANTs image class: from ANTsR to oro.nifti class, using the package extrantsr.

3.9 Preprocessing with ANTsR

  • Bias field correction (N4) Use function extrantsr:: bias_correct for correction
n3img = bias_correct(nim,correction = "N3", retimg = TRUE)
n4img = bias_correct(nim,correction = "N4", retimg = TURE)
  • Registration Use function ants_regwrite from package extrantsr to register the image to the template
reg_n4 = ants_regwrite(filename= n4img, template.file = template, remove.warp = TRUE, typeofTransform = "Rigid")

Theme 4: Registration, ROI quantification, segmentation

4.1 Co-registration

See section 3.2

4.2 fslr co-registration

Use flirt function from fslr package to register the T2w(infile) to the T1 (reffile)

4.3 AnTsR co-registration

Use ants_regwrite from extrantsr package to register the T2(filename) to the T1 (template.file).

4.4 wrapper functions

  • Use function preprocess_mri_within from extrantsr function to do the following steps:
    • Inhomogeneity Correction
    • Registration of the files to the first filename

4.5 Across-visit co-regisration of T1 images

  • Visit 1 files are in the visit 1 T1 space and visit 2 files are in the visit 2 T1 space.
  • Register the follow-up/visit 2 T1 scan to the visit 1/ baseline scan.

Overview

  • Registration within a subject
    • ants_regwrite wraps around the reading/writing of images and applying transformations
    • double_ortho and ortho2 can provide basic visual checks.
  • When images are registered in the space space, we can perform the following operations:
    • masking with a brain mask
    • transforming images to new spaces with one modality

4.6 Rigid Registration to T1

Overall Framework

FLAIR Image + Tumor ROI -> T1 weighted image -> Template Image

The first is rigid registration, the second is nonlinear registration.

4.7 Affine registration of T1 to Template

Use function ants_regwrite, where typeofTransform = "Affine".

4.8 Nonlinear registration of T1 to template

Use function ants_regwrite, where typeofTransform = "SyN".

SyN: symmetric image normalization, a symmetric diffeomorphic registration technique.

4.9 Getting ROI anatomic info from non-linear registration

  • Motivation: we have the ROI in the template space, and want to infer its location by leveraging the atlas labels.
  • Dealing with non-binary ROI
    • Threshold the ROI
    • Use a weighted sum over the ROI
  • Quantifiying the ROI engagement: tumor by region / region by Tumor

4.10 Tissue-Level Segmentation

We show how to use fslr to

  • segment CSF,GM, and WM
  • calculate the volume of CSF, GM, and WM.

Functions

  • readNIfTI
  • bias_correct
  • fslbet_robust: performs skull-stripping
  • fast: calls fast from FSL, does Segmentation
  • ortho2: does orthographic display

The function fast returns probability maps for corresponding tissue classes: pve_0, pve_1, pve_2

  • fast==1 represents CFS
  • fast==2 represents grey matter
  • fast==3 represents white matter

Reference

Imaging in R

Introduction to Neurohacking in R