MedOtter/UCSF-PDGM

UCSF-PDGM — BraTS-format structural subset (T1/T1c/T2/FLAIR + tumor segmentation)

*⚠️ This repository is the BraTS-format structural subset of UCSF-PDGM — NOT the full multi-sequence release. Per patient it provides the 4 structural MRI modalities (T1, T1c, T2, FLAIR) co-registered to the 240 × 240 × 155 BraTS grid (1 mm isotropic, skull-stripped) plus the expert-corrected tumor segmentation. The full UCSF-PDGM release additionally contains DWI, ADC, ASL, SWI, HARDI/DTI and derived diffusion maps, N4 bias-corrected (`_bias`) images, and whole-brain / parenchyma masks — all intentionally excluded here* to keep a compact, BraTS-compatible glioma-segmentation dataset.

The University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM) dataset is a cohort of preoperative 3 T brain MRI from patients with histopathologically confirmed diffuse gliomas.

Contents

• —495 patients — one folder each, UCSF-PDGM-XXXX/, with 5 NIfTI volumes.
• —UCSF-PDGM-metadata_v5.csv — official clinical metadata (sex, age, WHO CNS grade, final WHO-2021 diagnosis, MGMT status/index, 1p/19q, IDH, overall survival, extent of resection, and the BraTS21 ID / cohort columns).
• —UCSF-PDGM-metadata_glossary.csv — glossary for the metadata columns.

Note: the metadata table has 501 rows (imaging studies) while there are 495 image folders (unique patients); 6 follow-up/duplicate studies are not included as images.

Per-patient series layout (do not infer — this is the authoritative mapping)

Every UCSF-PDGM-XXXX/ folder contains exactly these five files:

All five volumes share identical geometry — shape (240, 240, 155), 1 mm isotropic — and are voxel-aligned.

Tumor segmentation labels (BraTS convention)

Masks were produced for the BraTS 2021 challenge: an automated ensemble of prior BraTS-winning algorithms, then manually corrected and approved by neuroradiologists.

⚠️ Benchmark-overlap warning (BraTS 2021 / 2023)

Roughly 298 of the 495 cases (~60 %) overlap with BraTS 2021, and therefore with BraTS 2023 (which inherits the BraTS 2021 glioma data). The BraTS21 ID, BraTS21 Segmentation Cohort, and BraTS21 MGMT Cohort columns in UCSF-PDGM-metadata_v5.csv identify exactly which cases. Do not treat UCSF-PDGM as an independent benchmark alongside BraTS2023 without first excluding the intersection — otherwise evaluation is contaminated by train/test leakage.

Joining metadata to images (ID padding differs)

Metadata IDs are 3-digit zero-padded (UCSF-PDGM-004); image folders/filenames are 4-digit (UCSF-PDGM-0004). Normalize by stripping the UCSF-PDGM- prefix and parsing the integer to join the two.

Preprocessing

• —Co-registered (nonlinear) to the T2-FLAIR reference space at 1 mm isotropic.
• —Skull-stripped with a deep-learning brain-extraction method.
• —Intensities are not N4 bias-corrected in this subset (the full release also offers _bias variants).

License

CC BY 4.0 (Creative Commons Attribution 4.0 International). You must give appropriate credit.

Citation

Calabrese E, Villanueva-Meyer JE, Rudie JD, et al. The University of California San Francisco Preoperative Diffuse Glioma MRI Dataset. Radiology: Artificial Intelligence. 2022;4(6):e220058. https://doi.org/10.1148/ryai.220058

Data (TCIA): Calabrese, E., et al. (2022). UCSF-PDGM (Version 5) [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/tcia.bdgf-8v37 — collection page: https://www.cancerimagingarchive.net/collection/ucsf-pdgm/

Provenance

Structural subset extracted from UCSF-PDGM Version 5 (TCIA), retaining only the T1/T1c/T2/FLAIR images and the tumor_segmentation mask from each patient's NIfTI package. Re-distributed here under the source CC BY 4.0 license.