2DeteCT - A large 2D expandable, trainable, experimental Computed Tomography dataset for machine learning.

Area of Science:

• Computational imaging
• Medical imaging
• Machine learning applications

Background:

• Machine learning (ML) for image reconstruction requires large datasets of measurements and ground-truth images.
• Experimental datasets for X-ray Computed Tomography (CT) are limited, hindering ML method development.
• Current methods often rely on simulated data, lacking real-world experimental validation.

Purpose of the Study:

• To provide a versatile, open-access 2D fan-beam CT dataset for the research community.
• To facilitate the development and evaluation of ML techniques for various image reconstruction tasks.
• To address the gap in experimental data for ML-based CT imaging.

Main Methods:

• Designed a semi-automatic scanning procedure using a flexible laboratory X-ray CT setup.
• Acquired 5,000 slices with diverse samples (varying shape and density) at high resolution.
• Collected data with three beam characteristics (high-fidelity, low-dose, beam-hardening) and 750 out-of-distribution slices.

Main Results:

• Generated a comprehensive 2D fan-beam CT dataset with raw projection data.
• Provided reference reconstructions and segmentations derived from an open-source processing pipeline.
• Dataset includes variations for robustness and segmentation tasks.

Conclusions:

• The open dataset enables advanced ML research in X-ray CT image reconstruction.
• Facilitates the development of more robust and accurate CT imaging techniques.
• Supports the validation of ML algorithms using realistic experimental data.