Skip to content

UCSF-PDGM – University of California San Francisco Preoperative Diffuse Glioma MRI

Choose a tag to compare

View all tags
@data-hcp data-hcp released this 20 Apr 14:35
· 8 commits to main since this release
ucsf-pdgm
a8140bc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

The UCSF-PDGM collection provides preoperative multi-parametric brain MRI and matched molecular, clinical, and follow-up data for adult patients with histopathologically confirmed WHO grade II–IV diffuse gliomas. All patients were scanned at the University of California San Francisco (UCSF) using a standardized 3T MRI protocol that emphasizes predominantly 3D acquisitions and includes advanced diffusion (HARDI) and perfusion (ASL) imaging.

In total, the dataset comprises 495 subjects (501 MRI exams) with harmonized mpMRI, tumor segmentations (aligned to BraTS standards), and curated IDH and MGMT biomarker status. This resource is designed to support AI and quantitative imaging research in areas such as automated tumor segmentation, radiogenomics, survival prediction, and treatment response modeling. All data are fully preoperative; prior tumor treatment is an exclusion criterion (biopsy allowed).

Overview

  • Dataset name: UCSF-PDGM – UCSF Preoperative Diffuse Glioma MRI
  • Institution: University of California San Francisco
  • Repository: The Cancer Imaging Archive (TCIA)
  • Species: Human
  • Subjects: 495 patients (501 exams)
  • Tumor types: WHO grade II–IV diffuse glioma
  • Data types: MRI (NIfTI), bval/bvec, tumor segmentations, clinical and molecular data
  • Total size: ~142 GB (imaging + annotations)
  • Latest version: Version 5 (updated 2025-05-30)
  • DOI: 10.7937/TCIA.BDGF-8V37
  • License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Study Population and Biomarkers

  • Population: Adult patients with histopathologically confirmed grade II–IV diffuse gliomas
  • Inclusion: Preoperative MRI, initial tumor resection, genetic testing at a single center (2015–2021)
  • Exclusion: Any prior brain tumor treatment (except biopsy)

Genetic biomarkers:

  • IDH mutation status available for all tumors
  • MGMT promoter methylation available for grade III–IV gliomas
  • 1p/19q codeletion reported for a subset of cases

Grade distribution (501 cases):

  • Grade II: 55 cases (11%)
  • Grade III: 42 cases (9%)
  • Grade IV: 403 cases (80%)

There is a consistent male predominance (~56–60%) across grades. IDH mutations are common in lower-grade gliomas (83% of grade II, 67% of grade III) and rare in grade IV (8%). MGMT hypermethylation is present in ~63% of grade IV gliomas.

Imaging Protocol

All MRIs were acquired preoperatively on a 3.0T GE Discovery 750 scanner using an 8-channel head coil. The standardized protocol includes:

  • Structural:

    • 3D T2-weighted
    • 3D T2/FLAIR-weighted
    • Susceptibility-weighted imaging (SWI)
    • Pre- and post-contrast T1-weighted (3D)

  • Diffusion:

    • 2D 55-direction HARDI diffusion sequence
    • Derived maps: DWI, FA, MD, AD, RD (via FSL Eddy + DTIFIT)

  • Perfusion:

    • 3D arterial spin labeling (ASL) perfusion imaging

Gadolinium-based contrast agents used:

  • Gadobutrol (Gadovist): 0.1 mL/kg
  • Gadoterate (Dotarem): 0.2 mL/kg

Image Pre-processing and Tumor Segmentation

Pre-processing:

  • HARDI data corrected with FSL Eddy (eddy current correction with outlier replacement; no topup)
  • Tensor fitting with FSL DTIFIT (simple least squares)
  • All contrasts registered and resampled to each subject’s T2/FLAIR space (1 mm isotropic) using ANTs non-linear registration
  • Skull stripping performed with a public deep-learning model:

Tumor segmentation:

  • Multicompartment segmentation performed as part of the BraTS 2021 pipeline
  • Initial automated segmentation using an ensemble of prior BraTS-winning models
  • Manual corrections by trained radiologists, with final approval by two expert reviewers
  • Segmented compartments:
    • Enhancing tumor
    • Non-enhancing / necrotic tumor
    • FLAIR hyperintense abnormality (“edema” region)

These labels support:

  • Benchmarking of segmentation algorithms
  • Radiomics and radiogenomics analyses
  • Survival and progression modeling

Data Access

All data are hosted on TCIA and are publicly accessible.

Version 5 changes (2025-05-30):

  • Fixed a header issue in DTI_eddy_noreg by providing NIfTI files in original orientation and spacing (post-FSL eddy, prior to further processing)
  • Added rotated bvecs for each exam (FSL eddy outputs)

Download resources:

Content Data Type Format Subjects License Access
Images & annotations MR images + segmentations NIfTI + BVEC 495 CC BY 4.0 Download via IBM Aspera (142 GB)
Clinical data Demographic, molecular, diagnosis, follow-up CSV 495 CC BY 4.0 Direct CSV download
bval files Diffusion b-values BVAL/ZIP — CC BY 4.0 Direct download
bvec files Rotated diffusion b-vectors BVEC/ZIP — CC BY 4.0 Direct download

Access details and download links are available on the TCIA collection page.

External Tools and Resources

Citation

Users must cite the dataset as:

Calabrese, E., Villanueva-Meyer, J., Rudie, J., Rauschecker, A., Baid, U., Bakas, S., Cha, S., Mongan, J., Hess, C. (2022).
The University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM) (Version 5) [dataset].
The Cancer Imaging Archive.
https://doi.org/10.7937/TCIA.BDGF-8V37

Usage Policy

The UCSF-PDGM collection is distributed under CC BY 4.0, allowing reuse and adaptation (including commercial use) with appropriate attribution.
Users must comply with the TCIA Data Usage Policy and Restrictions.

Source

© 2025 The Cancer Imaging Archive (TCIA).
Prepared for redistribution under data-others/disease/ucsf-pdgm by the Pittsburgh Fiber Data Hub.

Assets 4,513
Loading