BraTS Preprocessing

Installation

pip install git+https://github.com/johncolby/brats_preprocessing

Usage

import brats_preprocessing.brats_preprocessing as bp

zip_path = '/path/to/<accession>.zip'
model_path = '/path/to/model.Rdata'

mri = bp.tumor_study(zip_path=zip_path, model_path=model_path)

mri.setup()
mri.classify_series()
mri.preprocess()

Inputs

These tools expect an <accession>.zip file as input. The archive should contain a single study directory, which should in turn contain series directories, which should in turn contain DICOM files. For example, if I unzip 11111111.zip into a directory named 11111111, then the structure should look like:

+-11111111/
  |
  +-a8651ef7f70c8cba45e3141f4e3bef17/
  | |
  | +-06b758b1335988a09ee4cb7af3678b98/
  | | |
  | | +-ae01489d59eed67281e13df5ed992af0.dcm
  | | |
  | | +-ad08d6483cb7b02865b3b05b8003a170.dcm
  | | |
  | | +-... <additional files>
  | |
  | +-49e1e56731aac53ea1205e1f080b5c70/
  | | |
  | | +-268f45ba1d77340ff0fe7c4c283e6284.dcm
  | | |
  | | +-fb60f2e00e74a0ddc7e0f5bcaad68df6.dcm
  | | |
  | | +-... <additional files>
  | |
  | +-bfa267d60ff3e8873518773b934f99fc/
  | | |
  | | +-15bce104269e2c8f8cdf274845277c08.dcm
  | | |
  | | +-c081ec04dcade7a2643a62186374111a.dcm
  | | |
  | | +-... <additional files>
  ...

Alternatively, if air_download is installed, you may use the download() method to access the AIR API directly.

mri = bp.tumor_study(acc='11111111')
mri.setup()
mri.download(URL='https://air.<domain>.edu/api/', cred_path='/path/to/air_login.txt')

Series classification

A small but important step in processing clinical data involves identifying the desired input series. For a given application/context, this can be done easily by training a small classifier to identify series type (e.g. axial 3D T1, FLAIR, etc.) based on DICOM header data (study code, TR, TE, series description, etc.). We do that here by loading a custom pre-trained model.Rdata file. See URL for details.

Alternatively, series directory paths may be manually added like:

mri.add_paths(['/path/to/flair', 
               '/path/to/t1', 
               '/path/to/t1ce', 
               '/path/to/t2'])

Preprocessing pipeline

Implemented with nipype.

DICOM files are converted to NIfTI (dcm2niix).
T1 processing.
1. Skull strip to get T1_brain and T1_brain_mask (BET).
2. Register T1_brain to MNI template space to get T1_brain_mni and t1_brain_mni.mat (FLIRT).
3. Apply t1_brain_MNI.mat transform to T1_brain_mask (FLIRT).
4. Apply final mask in MNI space (fslmaths).
Non T1 processing, e.g. FLAIR.
1. Register to T1 but only save .mat (FLIRT).
2. Concatenate FLAIR-to-T1 and T1-to-MNI transformation matrices (convert_xfm).
3. Apply concatenated .mat to register FLAIR to MNI in a single resampling step (FLIRT).
4. Apply final mask in MNI space (fslmaths).
Multi-channel bias field correction (FAST).
Merge into single 4D file (fslmerge).
Change data type to 16 bit integer (fslmaths).
Pad voxel dimensions and reorient to BraTS template space (FLIRT).

Notes on the BraTS template

MNI and BraTS template systems are different.

	MNI	BraTS
Voxel size	1 x 1 x 1	1 x 1 x 1
Voxel dimensions	182 x 218 x 182	240 x 240 x 155
Origin	`[90, -126, -72]`	`[0, 239, 0]`
Orientation	LAS (radiological)	LPS (neurological)

Calculate voxel offsets needed to match origins between BraTS and MNI.

90  + (240 - 182) / 2 # x = 119
126 + (240 - 218) / 2 # y = 137
72  + (155 - 182) / 2 # z = 58.5

Can generate a minimal BraTS orientation template de novo like:

fslcreatehd 240 240 155 1 1 1 1 1 119 137 59 4 brats_ref
fslorient -setsformcode 1 brats_ref
fslorient -forceneurological brats_ref
fslswapdim brats_ref x -y z brats_ref_reorient

This is included as brats_ref_reorient.nii.gz

To convert from MNI to BraTS template space (pad voxel dimensions and reorient), we can use FSL FLIRT:

flirt -in t1_brain_flirt -ref brats_ref_reorient -applyxfm -usesqform -out output