CT Data of a Pen-Spring: Application to Under-Sampled Dynamic X-ray Tomography

Juliaana Juurakko, Zenith Purisha, Simo SärkkäarXiv 2019 · DOI · arXiv

TL;DR

This paper introduces an open-access Computed Tomography (CT) dataset of a common pen-spring, specifically designed to benchmark algorithms for under-sampled and dynamic X-ray reconstruction. By providing both raw sinograms and pre-computed measurement matrices, the authors offer a "plug-and-play" resource for researchers developing advanced imaging techniques for challenging applications like cardiac CT.

What Problem Does This Paper Address?

In the world of X-ray tomography, "dynamic" imaging—capturing an object in motion—is a major hurdle. To avoid motion blur, you need to take images very quickly, which often means taking fewer projections (under-sampling). However, reconstructing a high-quality 3D image from sparse, under-sampled data is an "ill-posed" mathematical problem that leads to significant artifacts.

While synthetic datasets (like the famous Shepp-Logan phantom) are common, they don't capture the messy realities of physics: beam hardening, detector noise, and subtle misalignments. There is a persistent need for real-world, accessible datasets that allow researchers to test how their algorithms (such as Kalman filters or sparse regularization) perform against actual hardware constraints.

Methods

The researchers used a Procon X-ray CTportable device to scan a 25mm long pen-spring. To simulate a dynamic process (like a spring moving through a 2D plane), they used a clever "stop-motion" approach:

  1. Geometry: They employed a fan-beam projection setup.
  2. The "Dynamic" Proxy: Instead of filming a vibrating spring, they lifted the spring vertically by 0.25mm between static scans. This allowed them to capture different cross-sections of the helical wire, which, when played in sequence, mimic the motion of a spring passing through a fixed 2D slice.
  3. Sampling: They captured a full set of 360 projections but down-sampled them to sets of 10 and 100 projections to test different levels of under-sampling.
  4. Processing: The raw data was binned, log-transformed, and normalized to produce a finalSino (sinogram) that fits the standard linear model $Ax = b$, where $A$ is the measurement matrix and $x$ is the desired reconstruction.

Datasets

The authors released the Pen-Spring CT Dataset on Zenodo. It is unique because it provides:

  • Multiple Resolutions: Data is provided for spatial resolutions of $64 \times 64$ and $256 \times 256$ pixels.
  • Temporal Instances: Depending on the set, there are 11 or 33 "time steps" (slices at different heights).
  • Ready-to-Use Matrices: Crucially, they include the measurement matrix $A$. Calculating this matrix is often the most computationally tedious part of CT research; providing it makes the dataset immediately accessible to mathematicians and computer scientists who may not be experts in X-ray physics.

Key Insights and Ideas

  • The Helix as a Test Case: A spring is an elegant test object. Its cross-section in a 2D slice is just a small dot (the wire) that moves in a circle as you move up the spring. This makes it very easy to visually track reconstruction errors or "lag" in dynamic algorithms.
  • Bridging the Gap: By providing the original measured sinograms alongside the processed ones, the authors allow researchers to either use the "clean" version for algorithm testing or the "raw" version to study pre-processing effects.
  • Real-World "Dirty" Data: Unlike a simulation, this data includes the physical footprint of a 0.5mm thick metal thread, providing a realistic challenge for edge-preserving reconstruction methods.

Key Results

The paper serves primarily as documentation for the dataset, but it demonstrates successful 2D slice reconstructions using 100 projections (see Figure 8 in the paper).

  • 10-Projection Set: Provides 33 different heights, offering higher "temporal" resolution but making the reconstruction much more difficult due to extreme under-sampling.
  • 100-Projection Set: Provides 11 different heights, offering a "denser" data set for testing higher-fidelity reconstructions.
  • The authors also provided a video (Reconstruction.mp4) showing the spring's cross-section moving through the slices, confirming the dataset's utility for dynamic modeling.

Limitations and Future Work

The primary limitation is that the "dynamics" are simulated via vertical displacement of a static object. This means the dataset does not capture mechanical deformation (the spring stretching or compressing), which would introduce different types of motion challenges.

The authors suggest that future work could involve stacking these sinograms to produce periodic data sets, further pushing the boundaries of how we reconstruct objects that move in predictable, repeating patterns—much like a beating human heart.

Summary updated

Abstract

This is the documentation of Computed Tomography (CT) data of a pen-spring. The open data set is available at https://zenodo.org/record/3266936#.XRyMdCZS9oA and can be freely used for scientific purposes with appropriate references to the data and to this document in arxiv.org. The provided data set includes the X-ray sinograms ({\tt finalSino}) of a single 2D slice from a different height of the spring. The {\tt finalSino} was obtained from a measured 10-projection or 100-projection {\tt sinogram} using fan-beam geometry by down-sampling and taking logarithms. The data set includes also those original measured {\tt sinogram}s and corresponding measurement matrices.
BibTeX
@misc{juurakko2019ct,
  author = {Juliaana Juurakko and Zenith Purisha and Simo Särkkä},
  title = {{CT Data of a Pen-Spring: Application to Under-Sampled Dynamic X-ray Tomography}},
  year = {2019},
  eprint = {1907.01871},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1907.01871},
}
Figure