Multicenter CT phantoms public dataset for radiomics reproducibility tests

Area of Science:

• Medical Imaging
• Radiomics
• Data Science

Background:

• Radiomics research requires reproducible features for reliable clinical prediction models.
• Ensuring the generalizability of radiomics models across different settings is crucial.
• Phantom data provides a controlled environment to assess feature stability.

Purpose of the Study:

• To describe a public, open-access computed tomography (CT) phantom image dataset.
• To facilitate radiomics reproducibility research by providing data acquired at multiple institutions.
• To enable testing of radiomic feature reproducibility across various acquisition parameters and scanner types.

Main Methods:

• Three phantoms (Catphan 700, COPDGene Phantom II, Triple modality 3D Abdominal Phantom) were scanned.
• Data were collected from three Dutch medical centers using Siemens and Philips scanners.
• Acquisition parameters including slice thickness, reconstruction kernels, and tube current were varied.

Main Results:

• A comprehensive CT phantom image dataset was created and made publicly available.
• The dataset is accessible via the Dutch "Extensible Neuroimaging Archive Toolkit-XNAT" (XNAT).
• The data is licensed under Creative Commons 3.0, allowing free reuse with attribution.

Conclusions:

• The dataset serves as a valuable resource for assessing CT radiomic feature reproducibility.
• It aids in identifying and excluding unstable features, enhancing the external validity of radiomics models.
• Future extensions aim to include other modalities and more complex phantom simulations.