Automated release for ucsd-ptgbm_4

Automated release for ucsd-ptgbm_3

Automated release for ucsd-ptgbm_2

Automated release for ucsd-ptgbm_1

This dataset provides a large-scale, post-operative glioblastoma MRI cohort with advanced imaging modalities, expert annotations, and clinical outcomes, designed to support research in tumor progression, treatment response, and radiogenomics.

• Source: https://www.cancerimagingarchive.net/collection/ucsd-ptgbm/
• DOI: https://doi.org/10.7937/fwv2-dt74

Dataset Overview

• Subjects: 178 patients

• Timepoints: 243 imaging sessions

• Cancer type: High-grade glioma (glioblastoma)

• Mean age: 56 ± 13 years

• Sex: 116 male, 62 female

• Total size: ~44.98 GB :contentReference[oaicite:0]{index=0}

• Scanner: 3T clinical MRI (GE systems)

Imaging Modalities

Structural MRI

• Pre- and post-contrast T1-weighted (3D IR-SPGR)
• T2-weighted FLAIR

Diffusion MRI (Multishell RSI)

• b-values: 0, 500, 1500, 4000 s/mm²
• Multi-direction acquisition
• Enables cellularity mapping via Restricted Spectrum Imaging (RSI)

Perfusion Imaging

• DSC (Dynamic Susceptibility Contrast)
• ASL (Arterial Spin Labeling)

Key Features

• Post-operative dataset (rare compared to pre-op datasets)

• Multishell diffusion MRI for separating:

  • Tumor cellularity
  • Edema / free water

• Expert-validated tumor segmentation

  • Based on BraTS standards
  • Manually refined by neuroradiologists

• Cellular tumor annotations

  • Enhancing tumor (ECT)
  • Non-enhancing tumor (NECT)
  • Total cellular tumor (TCT)

• Clinical and molecular data

  • IDH mutation status
  • MGMT promoter methylation
  • Overall survival (OS)
  • Progression-free survival (PFS)

Dataset Composition

Imaging Data

• Multimodal MRI (NIfTI format)
• Co-registered to 1 mm isotropic MNI space

Segmentations

• Multi-compartment tumor labels
• Radiologist-approved voxelwise annotations

Clinical Data

• Demographics
• Diagnosis and treatment
• Follow-up outcomes

Additional Files

• b-values / b-vectors
• Negative case categorization

Cohort Details

• Residual or recurrent tumor: 192 timepoints

• Post-treatment changes only: 51 timepoints

  • Includes:
    • Pseudoprogression
    • Radiation necrosis
    • Non-specific treatment effects

• Survival data available: subset of 94 patients

Preprocessing

• RSI cellularity modeling via linear mixture modeling
• Beam-forming filter for signal refinement
• DSC processed with leakage-corrected CBV
• Skull stripping via nnUNet-based pipeline
• All modalities registered to MNI space

Scientific Applications

• Post-treatment tumor assessment
• Differentiation of recurrence vs treatment effects
• Radiogenomics and biomarker discovery
• AI/ML for segmentation and prognosis
• Diffusion-based tumor microstructure modeling

License

• Creative Commons Attribution 4.0 (CC BY 4.0)

Citation

Gagnon et al., 2025
The UCSD-PTGBM dataset (Version 3)
https://doi.org/10.7937/fwv2-dt74

Notes

• One of the few publicly available post-operative glioma datasets
• Includes advanced diffusion + perfusion imaging, uncommon in open datasets
• Particularly valuable for clinical translation and longitudinal analysis

participants.tsv Fields

The participants.tsv file includes scanner, demographic, diagnostic, molecular, treatment, and outcome variables for each case.

The UPENN-GBM collection provides multi-parametric magnetic resonance imaging (mpMRI) scans and corresponding clinical, histopathologic, and radiomic data from 630 patients with de novo glioblastoma (GBM).
The dataset was curated by the University of Pennsylvania Health System and released through The Cancer Imaging Archive (TCIA) as a comprehensive, open-access imaging resource for studying glioblastoma biology, segmentation reproducibility, and radiogenomic biomarkers.

Each subject includes co-registered and skull-stripped mpMRI scans together with automated and manually corrected tumor segmentation labels that delineate histologically distinct subregions (enhancing core, necrotic core, edema, etc.). These segmentations were refined and approved by expert board-certified neuroradiologists, enabling quantitative analyses without repeated manual annotation.

The dataset also provides a large panel of radiomic features, clinical outcomes, and molecular data, supporting cross-disciplinary research linking imaging, histology, and genomics. For a subset of cases, matched H&E-stained whole-slide histopathology images from resected tumor tissue are available, enabling radiology–pathology correlation.

Overview

• Dataset name: UPENN-GBM (Multi-parametric MRI for De Novo Glioblastoma)
• Institution: University of Pennsylvania Health System
• Repository: The Cancer Imaging Archive (TCIA)
• Species: Human
• Subjects: 630 patients
• Cancer type: Glioblastoma multiforme (GBM)
• Data types: MRI (DICOM/NIfTI), segmentation labels, histopathology, demographics, molecular and radiomic features
• Total size: ~357 GB
• Version: 2 (Updated 2022-10-24)
• DOI: 10.7937/TCIA.709X-DN49
• License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Imaging and Data Modalities

Study Description

This dataset integrates clinical, imaging, and molecular data to enable large-scale computational and translational research in glioblastoma.
All MRI volumes were preprocessed (skull stripping and co-registration) prior to segmentation, which was performed via an automated pipeline followed by manual expert correction.
Derived features include:

• Intensity and histogram-based measures
• Volumetric and morphological statistics
• Textural parameters (GLCM, GLRLM, etc.)
• Radiomic descriptors consistent with CaPTk and IBSI standards

The dataset supports:

• Benchmarking of automated tumor segmentation algorithms
• Radiogenomic association studies linking imaging phenotypes to molecular subtypes
• Outcome prediction (e.g., overall survival, progression-free survival)
• Radiology–pathology correlation and cross-modality feature harmonization

Data Access

All data are publicly available through The Cancer Imaging Archive (TCIA).
Use of NBIA Data Retriever or IBM Aspera Connect is required for large downloads.

Version 2 Updates (October 2022):

• Added digitized histopathology (NDPI format)
• Added radiology–pathology mapping CSV
• Harmonized radiomic feature files and metadata

Citation

Bakas, S., Sako, C., Akbari, H., Bilello, M., Sotiras, A., Shukla, G., Rudie, J. D., Flores Santamaria, N., Fathi Kazerooni, A., Pati, S., Rathore, S., Mamourian, E., Ha, S. M., Parker, W., Doshi, J., Baid, U., Bergman, M., Binder, Z. A., Verma, R., … Davatzikos, C. (2021).
Multi-parametric magnetic resonance imaging (mpMRI) scans for de novo Glioblastoma (GBM) patients from the University of Pennsylvania Health System (UPENN-GBM) (Version 2).
The Cancer Imaging Archive.
https://doi.org/10.7937/TCIA.709X-DN49

Users must include this citation in all publications derived from this dataset.

Usage Policy

This collection is released under CC BY 4.0, allowing sharing and adaptation for any purpose (including commercial), provided appropriate attribution is given.
All users must comply with the TCIA Data Usage Policy and Restrictions.

Acknowledgments

This dataset was developed through the collaboration of the University of Pennsylvania Health System, The Cancer Imaging Archive (TCIA), and the National Cancer Institute’s Cancer Imaging Program (CIP).
We thank the contributing radiologists, data scientists, and patients for enabling open-access cancer imaging research.

External Resources

• TCIA Collection Page – UPENN-GBM
• Imaging Data Commons (IDC)
• CaPTk: Cancer Imaging Phenomics Toolkit

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

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:
  • https://github.com/ecalabr/brain_mask

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:

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

External Tools and Resources

• Skull stripping model:
  https://github.com/ecalabr/brain_mask/

• Related datasets and benchmarks:

  • RSNA-ASNR-MICCAI BraTS 2021 challenge dataset

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

• TCIA collection page – UCSF-PDGM:
  (search “UCSF-PDGM” on The Cancer Imaging Archive)
• About TCIA:
  https://www.cancerimagingarchive.net/

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