torchio is a Python package containing a set of tools to efficiently
read, sample and write 3D medical images in deep learning applications
written in PyTorch,
including intensity and spatial transforms
for data augmentation and preprocessing. Transforms include typical computer vision operations
such as random affine transformations and also domain-specific ones such as
simulation of intensity artifacts due to
MRI magnetic field inhomogeneity
or k-space motion artifacts.
This package has been greatly inspired by NiftyNet.
The best way to quickly understand and try the library is the Jupyter notebook hosted by Google Colab. It includes many examples and visualization of most of the classes and even training of a 3D U-Net for brain segmentation of T1-weighted MRI with whole images and patch-based sampling.
If you like this repository, please click on Star!
If you used this package for your research, please cite this repository using the information available on its Zenodo entry or use this text:
Pérez-García, Fernando. (2020, January 15). fepegar/torchio: TorchIO: Tools for loading, augmenting and writing 3D medical images on PyTorch. Zenodo. http://doi.org/10.5281/zenodo.3598622
BibTeX entry:
@software{perez_garcia_fernando_2020_3598622,
author = {Pérez-García, Fernando},
title = {{fepegar/torchio: TorchIO: Tools for loading,
augmenting and writing 3D medical images on
PyTorch}},
month = jan,
year = 2020,
publisher = {Zenodo},
doi = {10.5281/zenodo.3598622},
url = {https://doi.org/10.5281/zenodo.3598622}
}This package is on the Python Package Index (PyPI). To install the latest published version, just run the following command in a terminal:
$ pip install --upgrade torchioThe Information eXtraction from Images (IXI) dataset contains "nearly 600 MR images from normal, healthy subjects", including "T1, T2 and PD-weighted images, MRA images and Diffusion-weighted images (15 directions)".
The usage is very similar to torchvision.datasets:
import torchio
import torchvision
transforms = [
torchio.ToCanonical(), # to RAS
torchio.Resample((1, 1, 1)), # to 1 mm iso
]
ixi_dataset = torchio.datasets.IXI(
'path/to/ixi_root/',
modalities=('T1', 'T2'),
transform=torchvision.transforms.Compose(transforms),
download=True,
)
print('Number of subjects in dataset:', len(ixi_dataset)) # 577
sample_subject = ixi_dataset[0]
print('Keys in subject sample:', tuple(sample_subject.keys())) # ('T1', 'T2')
print('Shape of T1 data:', sample_subject['T1'][torchio.DATA].shape) # [1, 180, 268, 268]
print('Shape of T2 data:', sample_subject['T2'][torchio.DATA].shape) # [1, 241, 257, 188]This is the dataset used in the notebook. It is a tiny version of IXI, containing 566 T1-weighted brain MR images and their corresponding brain segmentations, all with size (83 x 44 x 55).
ImagesDataset is a reader of 3D medical images that directly inherits from
torch.utils.Dataset.
It can be used with a
torch.utils.DataLoader
for efficient loading and data augmentation.
It receives a list of subjects, where each subject is an instance of
torchio.Subject containing instances of
torchio.Image.
The file format must be compatible with NiBabel or
SimpleITK readers.
import torchio
from torchio import ImagesDataset, Image, Subject
subject_a = Subject([
Image('t1', '~/Dropbox/MRI/t1.nrrd', torchio.INTENSITY),
Image('label', '~/Dropbox/MRI/t1_seg.nii.gz', torchio.LABEL),
])
subject_b = Subject(
Image('t1', '/tmp/colin27_t1_tal_lin.nii.gz', torchio.INTENSITY),
Image('t2', '/tmp/colin27_t2_tal_lin.nii', torchio.INTENSITY),
Image('label', '/tmp/colin27_seg1.nii.gz', torchio.LABEL),
)
subjects_list = [subject_a, subject_b]
subjects_dataset = ImagesDataset(subjects_list)
subject_sample = subjects_dataset[0]torchio includes grid, uniform and label patch samplers. There is also an
aggregator used for dense predictions.
For more information about patch-based training, see
NiftyNet docs.
import torch
import torch.nn as nn
import torchio
CHANNELS_DIMENSION = 1
patch_overlap = 4
patch_size = 128
grid_sampler = torchio.inference.GridSampler(
input_data, # some PyTorch tensor or NumPy array
patch_size,
patch_overlap,
)
patch_loader = torch.utils.data.DataLoader(grid_sampler, batch_size=4)
aggregator = torchio.inference.GridAggregator(
input_data, # some PyTorch tensor or NumPy array
patch_overlap,
)
model = nn.Module()
model.to(device)
model.eval()
with torch.no_grad():
for patches_batch in patch_loader:
input_tensor = patches_batch['image'].to(device)
locations = patches_batch['location']
logits = model(input_tensor)
labels = logits.argmax(dim=CHANNELS_DIMENSION, keepdim=True)
outputs = labels
aggregator.add_batch(outputs, locations)
output_tensor = aggregator.get_output_tensor()A patches Queue (or buffer) can be used for randomized patch-based sampling
during training.
This interactive animation
can be used to understand how the queue works.
import torch
import torchio
patches_queue = torchio.Queue(
subjects_dataset=subjects_dataset, # instance of torchio.ImagesDataset
max_length=300,
samples_per_volume=10,
patch_size=96,
sampler_class=torchio.sampler.ImageSampler,
num_workers=4,
shuffle_subjects=True,
shuffle_patches=True,
)
patches_loader = DataLoader(patches_queue, batch_size=4)
num_epochs = 20
for epoch_index in range(num_epochs):
for patches_batch in patches_loader:
logits = model(patches_batch) # model is some torch.nn.ModuleThe transforms package should remind users of
torchvision.transforms.
They take as input the samples generated by an ImagesDataset.
A transform can be quickly applied to an image file using the command-line
tool torchio-transform:
$ torchio-transform input.nii.gz RandomMotion output.nii.gz --kwargs "proportion_to_augment=1 num_transforms=4"Magnetic resonance images suffer from motion artifacts when the subject moves during image acquisition. This transform follows Shaw et al., 2019 to simulate motion artifacts for data augmentation.
Discrete "ghost" artifacts may occur along the phase-encode direction whenever the position or signal intensity of imaged structures within the field-of-view vary or move in a regular (periodic) fashion. Pulsatile flow of blood or CSF, cardiac motion, and respiratory motion are the most important patient-related causes of ghost artifacts in clinical MR imaging (From mriquestions.com).
Also known as Herringbone artifact, crisscross artifact or corduroy artifact, it creates stripes in different directions in image space due to spikes in k-space.
MRI magnetic field inhomogeneity creates slow frequency intensity variations. This transform is very similar to the one in NiftyNet.
Randomly swaps patches in the image. This is typically used in context restoration for self-supervised learning.
Adds noise sampled from a normal distribution with mean 0 and standard
deviation sampled from a uniform distribution in the range std_range.
It is often used after ZNormalization, as the output of
this transform has zero-mean.
Blurs the image using a discrete Gaussian image filter.
Reverse the order of elements in an image along the given axes.
Random affine transformation of the image keeping center invariant.
Implementation of New variants of a method of MRI scale standardization adapted from NiftyNet.
Rescale intensity values in an image to a certain range.
This transform first extracts the values with intensity greater than the mean, which is an approximation of the foreground voxels. Then the foreground mean is subtracted from the image and it is divided by the foreground standard deviation.
Resample images to a new voxel spacing using nibabel.
Pad images, like in torchvision.transforms.Pad.
Crop images passing 1, 3, or 6 integers, as in Pad.
Reorder the data so that it is closest to canonical NIfTI (RAS+) orientation.
Crops or pads image center to a target size, modifying the affine accordingly.
Applies a user-defined function as transform.
For example, image intensity can be inverted with
Lambda(lambda x: -x, types_to_apply=[torchio.INTENSITY])
and a mask can be negated with
Lambda(lambda x: 1 - x, types_to_apply=[torchio.LABEL]).
This example shows the improvement in performance when multiple workers are used to load and preprocess the volumes using multiple workers.
import time
import multiprocessing as mp
from tqdm import trange
import torch.nn as nn
from torch.utils.data import DataLoader
from torchvision.transforms import Compose
from torchio import ImagesDataset, Queue, DATA
from torchio.data.sampler import ImageSampler
from torchio.utils import create_dummy_dataset
from torchio.transforms import (
ZNormalization,
RandomNoise,
RandomFlip,
RandomAffine,
)
# Define training and patches sampling parameters
num_epochs = 4
patch_size = 128
queue_length = 400
samples_per_volume = 10
batch_size = 4
class Network(nn.Module):
def __init__(self):
super().__init__()
self.conv = nn.Conv3d(
in_channels=1,
out_channels=3,
kernel_size=3,
)
def forward(self, x):
return self.conv(x)
model = Network()
# Create a dummy dataset in the temporary directory, for this example
subjects_list = create_dummy_dataset(
num_images=100,
size_range=(193, 229),
force=False,
)
# Each element of subjects_list is an instance of torchio.Subject:
# subject = Subject(
# torchio.Image('one_image', path_to_one_image, torchio.INTENSITY),
# torchio.Image('another_image', path_to_another_image, torchio.INTENSITY),
# torchio.Image('a_label', path_to_a_label, torchio.LABEL),
# )
# Define transforms for data normalization and augmentation
transforms = (
ZNormalization(),
RandomNoise(std_range=(0, 0.25)),
RandomAffine(scales=(0.9, 1.1), degrees=10),
RandomFlip(axes=(0,)),
)
transform = Compose(transforms)
subjects_dataset = ImagesDataset(subjects_list, transform)
# Run a benchmark for different numbers of workers
workers = range(mp.cpu_count() + 1)
for num_workers in workers:
print('Number of workers:', num_workers)
# Define the dataset as a queue of patches
queue_dataset = Queue(
subjects_dataset,
queue_length,
samples_per_volume,
patch_size,
ImageSampler,
num_workers=num_workers,
)
batch_loader = DataLoader(queue_dataset, batch_size=batch_size)
start = time.time()
for epoch_index in trange(num_epochs, leave=False):
for batch in batch_loader:
# The keys of batch have been defined in create_dummy_dataset()
inputs = batch['one_modality'][DATA]
targets = batch['segmentation'][DATA]
logits = model(inputs)
print('Time:', int(time.time() - start), 'seconds')
print()Output:
Number of workers: 0
Time: 394 seconds
Number of workers: 1
Time: 372 seconds
Number of workers: 2
Time: 278 seconds
Number of workers: 3
Time: 259 seconds
Number of workers: 4
Time: 242 seconds