A Whole-Body PSMA-PET/CT dataset with manually annotated tumor lesions.

Area of Science:

• Medical Imaging
• Radiochemistry
• Oncology

Background:

• Prostate cancer diagnosis and staging rely heavily on imaging techniques.
• Accurate segmentation of lesions in PET/CT is crucial for treatment planning.
• Large, annotated datasets are needed to develop advanced AI tools for medical image analysis.

Purpose of the Study:

• To introduce a comprehensive, publicly available dataset of whole-body PET/CT scans for prostate carcinoma.
• To facilitate the development of deep learning models for automated lesion detection and segmentation.
• To enable multi-tracer machine learning model development for enhanced PET/CT analysis.

Main Methods:

• Compiled a dataset of 597 whole-body PET/CT studies using PSMA-targeting radiotracers ([18F]PSMA and [68Ga]Ga-PSMA-11) from 378 patients.
• Acquired scans between 2014-2022 on clinical PET/CT scanners, covering skull base to mid-thigh.
• Manually segmented all PSMA-expressing tumor lesions in 3D space and provided anonymized DICOM files, segmentation masks, and a TSV file with metadata.

Main Results:

• The dataset includes 597 whole-body PET/CT studies with detailed annotations.
• It contains anonymized DICOM images, 3D segmentation masks, and patient/scan metadata.
• Demonstrated the utility of the dataset for deep learning-based automated PET/CT analysis.

Conclusions:

• The described dataset is a valuable resource for advancing automated analysis of PSMA PET/CT in prostate cancer.
• It supports the development of AI models for improved lesion segmentation and multi-tracer analysis.
• Availability of this dataset aids research in medical image computing and computer-assisted interventions.