A cone-beam X-ray computed tomography data collection designed for machine learning.

Area of Science:

• Medical Imaging
• Computational Science
• Materials Science

Background:

• X-ray cone-beam (CB) computed tomography (CT) is crucial for detailed imaging.
• High cone-angle artifacts limit the quality and interpretability of CBCT images.
• Existing datasets often lack the specific design for machine learning and artifact reduction.

Purpose of the Study:

• To present a novel, open-access X-ray CBCT dataset tailored for machine learning.
• To facilitate research in high cone-angle artifact reduction in CBCT.
• To provide a comprehensive resource for developing and evaluating various image processing algorithms.

Main Methods:

• Acquired CBCT projection data from 42 walnuts using a laboratory X-ray setup.
• Utilized three different source orbits to capture varying cone angles and enable ground truth generation.
• Provided a complete image reconstruction pipeline including raw data, geometry, scripts, and reconstructed volumes.

Main Results:

• Generated a diverse dataset of X-ray CBCT scans from naturally variable objects (walnuts).
• Created artifact-free, high-quality ground truth images essential for supervised learning.
• The dataset supports artifact reduction, limited-angle reconstruction, super-resolution, and segmentation tasks.

Conclusions:

• The released dataset is a valuable resource for advancing machine learning in X-ray CBCT.
• It enables significant improvements in high cone-angle artifact reduction and other image reconstruction challenges.
• This open data promotes further research and development in medical and industrial imaging applications.