A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions

Area of Science:

• Medical Imaging
• Radiology
• Oncology

Background:

• Positron Emission Tomography/Computed Tomography (PET/CT) is crucial for cancer diagnosis and staging.
• Annotated datasets are essential for developing and validating AI tools in medical imaging.
• Publicly available data facilitates research reproducibility and accelerates advancements.

Purpose of the Study:

• To present a comprehensive, publicly accessible dataset of annotated whole-body Fluorodeoxyglucose (FDG)-PET/CT studies.
• To provide annotated malignant lesions for training and evaluating AI algorithms.
• To demonstrate the utility of the dataset for deep learning-based automated analysis.

Main Methods:

• Compilation of 1014 whole-body FDG-PET/CT datasets (501 cancer patients, 513 controls) acquired between 2014-2018.
• Manual, slice-per-slice 3D segmentation of all identified malignant FDG-avid lesions.
• Provision of anonymized DICOM files, segmentation masks, and image processing scripts.

Main Results:

• A large-scale, annotated FDG-PET/CT dataset is now publicly available.
• The dataset includes original DICOM images, 3D segmentation masks, and non-imaging data.
• A trained deep learning model for automated PET/CT analysis is provided.

Conclusions:

• This dataset serves as a valuable resource for advancing automated analysis of PET/CT imaging.
• It supports the development of AI-driven tools for improved cancer detection and characterization.
• The availability of annotated data promotes innovation in medical AI research.