LumbarSR: A Paired Clinical CT and Photon-Counting Micro-CT Dataset for Human Lumbar Vertebrae

Area of Science:

• Medical Imaging
• Bone Biology
• Computational Anatomy

Background:

• Low back pain affects millions, often linked to lumbar spine degeneration.
• Trabecular bone microarchitecture changes contribute to pain but are unresolved by standard clinical CT due to large voxel sizes.
• Current CT technology limits visualization of crucial trabecular structures (100-200 μm) with typical resolutions (500-1000 μm).

Purpose of the Study:

• Introduce LumbarSR, a novel dataset for developing and evaluating super-resolution (SR) methods for lumbar vertebra CT.
• Enable detailed analysis of trabecular bone structure invisible on routine clinical CT.
• Facilitate research into the link between bone microarchitecture and low back pain.

Main Methods:

• Paired dataset of 30 human lumbar vertebral specimens.
• Scanned using photon-counting micro-CT (Micro-PCCT) at 105 μm resolution as reference.
• Clinical CT scans acquired under 8 configurations (varying resolution, slice thickness, kernels) and registered to Micro-PCCT using ANTs.
• Data provided in DICOM and NIfTI formats.

Main Results:

• Established a benchmark dataset (LumbarSR) with registered high-resolution Micro-PCCT and standard clinical CT data.
• Performed baseline evaluations of image quality, trabecular morphometry, and SR method performance.
• Demonstrated the feasibility of voxel-wise evaluation and supervised learning for SR CT of lumbar vertebrae.

Conclusions:

• LumbarSR provides a valuable resource for advancing super-resolution CT techniques for lumbar spine analysis.
• The dataset supports the development of methods to visualize and quantify trabecular bone structure relevant to low back pain.
• Future research can leverage LumbarSR for improved understanding of degenerative spinal conditions.