Source-linked AI summary
The University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM) Dataset
Evan Calabrese, Javier E. Villanueva-Meyer, Jeffrey D. Rudie, Andreas M. Rauschecker, Ujjwal Baid, Spyridon Bakas, Soonmee Cha, John T. Mongan, Christopher P. Hess
TL;DR
Publicly available glioma MRI datasets have supported AI research but have largely been limited to four MRI contrasts. The UCSF-PDGM dataset addresses this gap with a standardized preoperative MRI resource combining advanced imaging, tumor segmentations, genetic data, and clinical outcomes for 500 patients.
Problem
Existing publicly available glioma MRI datasets have supported AI research but have largely been limited to four MRI contrasts.
Method
The study assembled 500 adults with histopathologically confirmed grade 2–4 diffuse gliomas imaged preoperatively at one medical center, with standardized MRI processing and multicompartment tumor segmentation.
Results
The UCSF-PDGM dataset provides standardized 3 Tesla 3D preoperative MRI, diffusion and perfusion imaging, multicompartment tumor segmentations, tumor genetic data, and treatment and survival data.
Takeaways & Limitations
Its 3D sequences and advanced MRI techniques, including ASL and HARDI, provide a new opportunity to explore imaging for AI applications in diffuse gliomas.
Abstract
from arXiv · showhide
Here we present the University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM) dataset. The UCSF-PDGM dataset includes 500 subjects with histopathologically-proven diffuse gliomas who were imaged with a standardized 3 Tesla preoperative brain tumor MRI protocol featuring predominantly 3D imaging, as well as advanced diffusion and perfusion imaging techniques. The dataset also includes isocitrate dehydrogenase (IDH) mutation status for all cases and O6-methylguanine-DNA methyltransferase (MGMT) promotor methylation status for World Health Organization (WHO) grade III and IV gliomas. The UCSF-PDGM has been made publicly available in the hopes that researchers around the world will use these data to continue to push the boundaries of AI applications for diffuse gliomas.
1 Introduction
Public glioma MRI datasets have enabled AI research, but existing resources were limited in MRI contrast diversity. The UCSF-PDGM dataset addresses this need with broader imaging data for diffuse gliomas.
- Existing public glioma MRI datasets have supported automated tumor segmentation, radiogenomics, and survival prediction.
- Prior publicly available datasets were largely limited to four MRI contrasts.
- The motivation for UCSF-PDGM is to expand publicly available imaging resources for AI research on diffuse gliomas.
2 Methods
The dataset was assembled from adults with confirmed grade 2–4 diffuse gliomas treated at one center, using standardized MRI, genetic testing, preprocessing, and expert-reviewed tumor segmentation.
- Patient population: The cohort included 500 adults with histopathologically confirmed grade 2–4 diffuse gliomas imaged before initial resection at one medical center from 2015 to 2021.Patients with prior brain tumor treatment were excluded, although prior biopsy was allowed.
- Genetic testing: IDH mutations were assessed by tissue sequencing, while all grade 3 and 4 tumors underwent MGMT methylation testing.
- MRI acquisition: Preoperative MRI used a 3.0 tesla scanner and included 3D structural, susceptibility, diffusion, perfusion, and contrast-enhanced sequences.The protocol included ASL perfusion and 55-direction HARDI imaging.
- Image preprocessing: HARDI data were corrected and converted into diffusion maps, while all contrasts were nonlinearly registered, resampled to 1 mm isotropic space, and skull stripped.
- Tumor segmentation: Tumor segmentations were generated with an automated ensemble, manually corrected, and approved by neuroradiologists across enhancing, necrotic, and FLAIR-abnormal compartments.
3 Results
The UCSF-PDGM cohort contains 500 grade 2–4 diffuse gliomas with genetic, survival, surgical, and multimodal MRI data. Its resources include 11 co-registered MRI contrasts, multicompartment segmentations, and public availability through challenge and archive platforms.
- Study participant demographics: MGMT promoter hypermethylation occurred in 63% of grade 4 gliomas, while 1p/19q codeletion occurred in 20% of grade 2, 5% of grade 3, and <1% of grade 4 tumors.
- Surgical treatment and survival: Overall survival data were included for the cohort, with glioblastoma survival stratified by gross total versus subtotal resection.
- MR image data: Each subject includes skull-stripped, co-registered 3D images in 11 MRI contrasts and multicompartment tumor segmentations.The representative images include structural, diffusion, susceptibility, perfusion, and post-contrast modalities.
- Comparison to related datasets: Compared with related public datasets, UCSF-PDGM provides a higher case count, a consistent 3 tesla protocol, and more sequences.
- Data availability: A portion was available through the 2021 BraTS challenge, with the complete dataset planned for public release through TCIA.
4 Discussion
UCSF-PDGM expands public diffuse glioma MRI resources for AI research by combining more cases with standardized 3D and advanced imaging. Its value depends on researchers using the dataset, particularly alongside existing resources.
- UCSF-PDGM increases the number of publicly available diffuse glioma MRI cases and adds 3D, ASL, and HARDI imaging.
- The dataset provides an opportunity to study the utility of advanced imaging techniques in AI applications for diffuse gliomas.
- Combining UCSF-PDGM with existing public datasets could support the next phase of radiologic AI research on diffuse gliomas.
- The dataset’s potential will be realized only if the radiology AI research community uses this resource.
5 Summary Statement
The UCSF-PDGM is a dataset of 500 patients with grade 2-4 diffuse gliomas. It combines standardized preoperative MRI with clinical, genetic, segmentation, treatment, and survival data.
- 500 patients with grade 2-4 diffuse gliomas comprise the UCSF-PDGM dataset.
- The dataset uses a standardized 3 Tesla, 3-dimensional, preoperative MR imaging protocol.
- Diffusion and perfusion MRI are provided alongside multicompartment tumor segmentations, tumor genetic data, and treatment/survival data.