We provide a large, annotated dataset of 597 whole-body PSMA-PET/CT studies from 378 male patients with suspected or diagnosed prostate carcinoma. Scans were acquired at LMU University Hospital, LMU Munich (2014–2022) on three clinical PET/CT scanners, jointly operated by the Departments of Nuclear Medicine and Radiology. All PSMA-avid tumor lesions were manually segmented on the PET images in 3D space using a dedicated software solution. 537 studies contain at least one lesion, while 60 show no lesions. The dataset includes anonymized DICOM files, DICOM segmentation masks, and a TSV file with patient age at imaging, PET/CT manufacturer and model name, PET radionuclide, and use of CT contrast agent. This dataset was used in the autoPET III and IV Grand Challenges to enable the development of machine-learning models for automated lesion segmentation in whole-body PET/CT.

Introduction

We provide a large, annotated dataset of whole-body PSMA-PET/CT studies from patients with suspected or diagnosed prostate cancer to support developing and benchmarking machine learning (ML) models for automated quantitative PET/CT analysis. Alongside the FDG-PET/CT dataset on TCIA , this dataset addresses the scarcity of publicly available, high-quality annotated PET/CT data. The FDG and PSMA-PET/CT datasets were jointly provided as training data for developing ML models in the autoPET III and autoPET IV Grand challenges for automated lesion segmentation in whole-body PET/CT.

Methods

Subject Inclusion and Exclusion Criteria

The institutional review board (Ethics Committee, Medical Faculty, LMU Munich), as well as the institutional data security and privacy review board, approved the publication of anonymized data. This retrospective dataset comprises 597 whole-body PSMA-PET/CT studies from 378 male patients (ages 48–92*) with suspected or diagnosed prostate carcinoma.
*Due to PHI criteria, all ages above 89 years in the metadata and the spreadsheet are set to 90 years, regardless of actual age.

Data Acquisition

Scans were conducted at LMU University Hospital, LMU Munich, between 2014 and 2022 using three clinical PET/CT scanners: Siemens Biograph mCT Flow 20, Siemens Biograph 64-4R TruePoint, and GE Discovery 690. 537 studies contain at least one PSMA-avid tumor lesion, 60 studies do not contain any PSMA-avid tumor lesion. The imaging protocol consisted of a diagnostic CT scan usually from the skull base to the mid-thigh with the following scan parameters: reference tube current exposure time product of 143 mAs (mean); tube voltage of 120 kV or 100 kV for most cases (range: [80, 140] kV), slice thickness of 2.5 - 5.0 mm (mean: 2.82 mm), and x-y resolution of mainly 0.98 mm. Intravenous contrast enhancement was used in most studies, except for patients with contraindications (26 studies). The whole-body PSMA-PET scan was acquired on average 74 minutes after intravenous injection of 246 MBq 18F-PSMA (mean, 369 studies) or 214 MBq 68Ga-PSMA (mean, 228 studies), respectively. The PET data was reconstructed with attenuation correction derived from corresponding CT data using standard, vendor-provided image reconstruction algorithms with a slice thickness ranging from 3.0 - 5.0 mm (mean: 3.49 mm) and x-y resolution ranging from 2.73 - 4.07 mm (mean: 3.56 mm).

Data Analysis

All PSMA-avid tumor lesions, including the primary tumor and/or all metastases, were manually segmented on the PET images by a single reader with 3 years of experience in hybrid imaging using dedicated software (mint Medical, Heidelberg, Germany) and validated by board-certified medical imaging experts with 4 years and >10 years of experience in hybrid imaging. Tumor lesions with significantly increased PSMA expression were segmented in 3D space by drawing circular VOIs, in which voxels with uptake values above a user-defined threshold were pre-segmented automatically and then manually corrected slice by slice, resulting in 3D binary segmentation masks and saved as NRRD files by the software. These files were exported and combined to a single segmentation mask per study and converted to DICOM SEG using the highdicom package v0.22.0 in Python v3.8.13. In addition, patient metadata was extracted from imaging DICOM tags: patient age at imaging (in years), PET/CT manufacturer and model name, PET radionuclide, and use of CT contrast agent. Information on radionuclides and the use of CT contrast agents was visually reviewed and validated by a radiologist with 10 years of experience in hybrid imaging.

Usage Notes

For each of the 597 PSMA PET/CT studies, we provide the anonymized original PET and CT DICOM files, and the corresponding segmentation mask as DICOM SEG. To view the DICOM data, we recommend open-source medical image data viewers such as 3D Slicer or the Medical Imaging Interaction Toolkit. For computational analysis, e.g. in Python, 3D image volumes can be read using open-source libraries such as pydicom, nibabel, or SimpleITK.

The patient metadata extracted from DICOM tags is shared in a TSV file. Each row contains the information on one study. Each study is uniquely identified by a case identifier number and the study date.

This dataset contains images of the head which, in theory, could pose re-identification risks using advanced image processing techniques. For this dataset TCIA implemented a “de-facing” pipeline to generate a version of this dataset without identifiable facial features which are published under an open-access license on TCIA. The original unaltered data will be made available at https://general.datacommons.cancer.gov/.

The Milli-focus X-ray Tube market, particularly focusing on models with a focal spot size of 0.4-1 mm, is an integral sector within the broader medical imaging and diagnostic equipment industry. These specialized X-ray tubes are essential for applications requiring high-resolution imaging, such as mammography and de

Introduction

MRI-based artificial intelligence (AI) research on patients with brain gliomas has been rapidly increasing in popularity in recent years in part due to a growing number of publicly available MRI datasets. Notable examples include The Cancer Genome Atlas Glioblastoma dataset (TCGA-GBM) consisting of 262 subjects and the International Brain Tumor Segmentation (BraTS) challenge dataset consisting of 542 subjects (including 243 preoperative cases from TCGA-GBM). The public availability of these glioma MRI datasets has fostered the growth of numerous emerging AI techniques including automated tumor segmentation, radiogenomics, and MRI-based survival prediction. Despite these advances, existing publicly available glioma MRI datasets have been largely limited to only 4 MRI contrasts (T2, T2/FLAIR, and T1 pre- and post-contrast) and imaging protocols vary significantly in terms of magnetic field strength and acquisition parameters. Here we present the University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM) dataset. The UCSF-PDGM dataset includes 501 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 O[6]-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.

Methods

Patient Population

Data collection was performed in accordance with relevant guidelines and regulations and was approved by the University of California San Francisco institutional review board with a waiver for consent. The dataset population consisted of 501* adult patients with histopathologically confirmed grade II-IV diffuse gliomas who underwent preoperative MRI, initial tumor resection, and tumor genetic testing at a single medical center between 2015 and 2021. Patients with any prior history of brain tumor treatment were excluded; however, history of tumor biopsy was not considered an exclusion criterion.

Genetic Biomarker Testing

All subjects’ tumors were tested for IDH mutations by genetic sequencing of tissue acquired during biopsy or resection. All grade III and IV tumors were tested for MGMT methylation status using a methylation sensitive quantitative PCR assay.

Study participant demographic data

The 501* cases included in the UCSF-PDGM include 55 (11%) grade II, 42 (9%) grade III, and 403 (80%) grade IV tumors. There was a male predominance for all tumor grades (56%, 60%, and 60%, respectively for grades II-IV). IDH mutations were identified in a majority of grade II (83%) and grade III (67%) tumors and a small minority of grade IV tumors (8%). MGMT promoter hypermethylation was detected in 63% of grade IV gliomas and was not tested for in a majority of lower grade gliomas. 1p/19q codeletion was detected in 20% of grade II tumors and a small minority of grade III (5%) and IV (<1%) tumors. Tabulated details and glossary are available in the Data Access and Detailed Description tabs below.

Image Acquisition

All preoperative MRI was performed on a 3.0 tesla scanner (Discovery 750, GE Healthcare, Waukesha, Wisconsin, USA) and a dedicated 8-channel head coil (Invivo, Gainesville, Florida, USA). The imaging protocol included 3D T2-weighted, T2/FLAIR-weighted, susceptibility-weighted (SWI), diffusion-weighted (DWI), pre- and post-contrast T1-weighted images, 3D arterial spin labeling (ASL) perfusion images, and 2D 55-direction high angular resolution diffusion imaging (HARDI). Over the study period, two gadolinium-based contrast agents were used: gadobutrol (Gadovist, Bayer, LOC) at a dose of 0.1 mL/kg and gadoterate (Dotarem, Guerbet, Aulnay-sous-Bois, France) at a dose of 0.2 mL/kg.

Image Pre-Processing

HARDI data were eddy current corrected and processed using the Eddy and DTIFIT modules from FSL 6.0.2 yielding isotropic diffusion weighted images (DWI) and several quantitative diffusivity maps: mean diffusivity (MD), axial diffusivity (AD), radial diffusivity (RD), and fractional anisotropy (FA). Eddy correction was performed with outlier replacement on and topup correction off. DTIFIT was performed with simple least squares regression. Each image contrast was registered and resampled to the 3D space defined by the T2/FLAIR image (1 mm isotropic resolution) using automated non-linear registration (Advanced Normalization Tools). Resampled co-registered data were then skull stripped using a previously described and publicly available deep-learning algorithm: https://www.github.com/ecalabr/brain_mask/.

Tumor Segmentation

Multicompartment tumor segmentation of study data was undertaken as part of the 2021 BraTS challenge. Briefly, image data first underwent automated segmentation using an ensemble model consisting of prior BraTS challenge winning segmentation algorithms. Images were then manually corrected by trained radiologists and approved by 2 expert reviewers. Segmentation included three major tumor compartments: enhancing tumor, non-enhancing/necrotic tumor, and surrounding FLAIR abnormality (sometimes referred to as edema).

The UCSF-PDGM adds to on an existing body of publicly available diffuse glioma MRI datasets that are commonly used in AI research applications. As MRI-based AI research applications continue to grow, new data are needed to foster development of new techniques and increase the generalizability of existing algorithms. The UCSF-PDGM not only significantly increases the total number of publicly available diffuse glioma MRI cases, but also provides a unique contribution in terms of MRI technique. The inclusion of 3D sequences and advanced MRI techniques like ASL and HARDI provides a new opportunity for researchers to explore the potential utility of cutting-edge clinical diagnostics for AI applications. In addition, these advanced imaging techniques may prove useful for radiogenomic studies focused on identification of IDH mutations or MGMT promoter methylation.

The UCSF-PDGM dataset, particularly when combined with existing publicly available datasets, has the potential to fuel the next phase of radiologic AI research on diffuse gliomas. However, the UCSF-PDGM dataset’s potential will only be realized if the radiology AI research community takes advantage of this new data resource. We hope that this dataset sparks inspiration in the next generation of AI researchers, and we look forward to the new techniques and discoveries that the UCSF-PDGM will generate.