BrainWebMRI#

class deepinv.datasets.BrainWebMRI(subject_ids=None, contrast='T1', download=True, root=None, transform=None)[source]#

Bases: ImageDataset

Dataset for BrainWeb.

BrainWeb brain phantom for Magnetic Resonance Imaging (MRI) research [1]. The dataset consists of 22 MRI brain phantom scans: 21 normal brains and 1 multiple sclerosis brain (patient 1). Several contrasts are available for each patient: T1, T2, T2* and PD. Each T1 volume has shape (1, 181, 256, 256) and is scaled by 1 / 4095 to that it is normalized with values in [0, 1].

This dataset relies on the original implementation of Pierre-Antoine Comby: <paquiteau/brainweb-dl>`_. Install it with pip install brainweb-dl.

Note

For a version of this dataset with dedicated features for emission tomography, such as emission / attenuation maps and hot lesions, see deepinv.datasets.BrainWebPET.

Parameters:

  • subject_ids (int, collections.abc.Sequence[int], None) – Subjects to include in the dataset. Possible values: [4, 5, 6, 18, 20, 38, 41-54]. Defaults to all subjects.

  • contrast (str) – MRI contrast to return: "T1", "T2", "T2*" or "PD". Defaults to "T1".

  • download (bool) – Download missing subjects. Defaults to True.

  • transform (collections.abc.Callable, None) – Optional volume transform.

  • root (str, pathlib.Path, None) – Root directory of dataset. Directory path from where we load and save the dataset.

References: