A cone-beam photon-counting CT dataset for spectral image reconstruction and deep learning

Area of Science:

• Medical Imaging
• Physics
• Computer Science

Background:

• Photon-counting CT (PCCT) is an emerging imaging technology with growing interest.
• Limited availability of real-world spectral PCCT datasets hinders the development and validation of advanced algorithms.
• Current research often relies on simulated data, potentially limiting real-world applicability.

Purpose of the Study:

• To introduce a novel, comprehensive dataset of real spectral photon-counting CT data.
• To facilitate research in spectral CT reconstruction, material decomposition, and deep learning applications.
• To enable fair and reproducible comparisons of image processing algorithms.

Main Methods:

• Acquisition of a cone-beam PCCT dataset using a custom micro-PCCT system.
• Scanning of 15 walnut samples across four bed positions with dual energy thresholds (15 keV and 30 keV).
• Provision of raw multi-energy projections, system parameters, calibration data, and reconstruction code.

Main Results:

• A dataset comprising 172,800 raw projection images (2063 × 505 pixels) is now publicly available.
• The dataset includes essential components for spectral CT studies, such as raw projections and system parameters.
• The data supports a wide range of applications, from basic reconstruction to complex deep learning models.

Conclusions:

• The release of this real PCCT dataset significantly alleviates the scarcity of such resources.
• This dataset will accelerate the development and validation of data-driven methods in spectral CT.
• It fosters advancements in medical imaging by promoting reproducible research and algorithm benchmarking.