brainweb 1.6.2

BrainWeb-based multimodal models of 20 normal brains

The following example may be launched interactively via any of the following:

• 

• Local file

• GitHub Preview

BrainWeb-based multimodal models of 20 normal brains

This project was initially inspired by “BrainWeb: 20 Anatomical Models of 20 Normal Brains.”

However there are a number of generally useful tools, image processing & display functions included in this project. For example, this includes volshow() for interactive comparison of multiple 3D volumes, get_file() for caching data URLs, and register() for image coregistration.

Download and Preprocessing for PET-MR Simulations

This notebook will not re-download/re-process files if they already exist.

• Output data

  • ~/.brainweb/subject_*.npz: dtype(shape): float32(127, 344, 344)

• Raw data source

  • ~/.brainweb/subject_*.bin.gz: dtype(shape): uint16(362, 434, 362)

• Install

  • pip install brainweb

---

• Author: Casper da Costa-Luis <[e-mail]>

• Date: 2017-2020

• Licence: MPLv2.0

from __future__ import print_function, division
%matplotlib notebook
import brainweb
from brainweb import volshow
import numpy as np
from os import path
from tqdm.auto import tqdm
import logging
logging.basicConfig(level=logging.INFO)

Raw Data

# download
files = brainweb.get_files()

# read last file
data = brainweb.load_file(files[-1])

# show last subject
print(files[-1])
volshow(data, cmaps=['gist_ncar']);

~/.brainweb/subject_54.bin.gz

Transform

Convert raw image data:

• Siemens Biograph mMR resolution (~2mm) & dimensions (127, 344, 344)

• PET/T1/T2/uMap intensities

  • PET defaults to FDG intensity ratios; could use e.g. Amyloid instead

• randomised structure for PET/T1/T2

  • t (1 + g [2 G_sigma(r) - 1]), where

    • r = rand(127, 344, 344) in [0, 1),

    • Gaussian smoothing sigma = 1,

    • g = 1 for PET; 0.75 for MR, and

    • t = the PET or MR piecewise constant phantom

# show region probability masks
PetClass = brainweb.FDG
label_probs = brainweb.get_label_probabilities(files[-1], labels=PetClass.all_labels)
volshow(label_probs[brainweb.trim_zeros_ROI(label_probs)], titles=PetClass.all_labels, frameon=False);

brainweb.seed(1337)

for f in tqdm(files, desc="mMR ground truths", unit="subject"):
    vol = brainweb.get_mmr_fromfile(
        f,
        petNoise=1, t1Noise=0.75, t2Noise=0.75,
        petSigma=1, t1Sigma=1, t2Sigma=1,
        PetClass=PetClass)

# show last subject
print(f)
volshow([vol['PET' ][:, 100:-100, 100:-100],
         vol['uMap'][:, 100:-100, 100:-100],
         vol['T1'  ][:, 100:-100, 100:-100],
         vol['T2'  ][:, 100:-100, 100:-100]],
        cmaps=['hot', 'bone', 'Greys_r', 'Greys_r'],
        titles=["PET", "uMap", "T1", "T2"],
        frameon=False);

~/.brainweb/subject_54.bin.gz

# add some lesions
brainweb.seed(1337)
im3d = brainweb.add_lesions(vol['PET'])
volshow(im3d[:, 100:-100, 100:-100], cmaps=['hot']);

# bonus: use brute-force registration to transform
#!pip install -U 'brainweb[register]'
reg = brainweb.register(
    data[:, ::-1], target=vol['PET'],
    src_resolution=brainweb.Res.brainweb,
    target_resolution=brainweb.Res.mMR)

volshow({
    "PET":    vol['PET'][:, 100:-100, 100:-100],
    "RawReg": reg[       :, 100:-100, 100:-100],
    "T1":     vol['T1' ][:, 100:-100, 100:-100],
}, cmaps=['hot', 'gist_ncar', 'Greys_r'], ncols=3, tight_layout=5, figsize=(9.5, 3.5), frameon=False);