Dynamic 3D Tomographic X-ray Data of Ladybug

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

TL;DR

Researchers at Aalto University have released an open-access dataset featuring dynamic 3D X-ray tomographic (CT) scans of a ladybug. This dataset provides a high-quality, real-world testbed for developing reconstruction algorithms that can handle moving objects—a critical challenge in medical imaging where patients are rarely perfectly still.

What Problem Does This Paper Address?

In the world of Computed Tomography (CT), the "gold standard" math assumes the object being scanned is perfectly stationary. However, in clinical reality, patients breathe, hearts beat, and people fidget. These movements create "motion artifacts" that can blur images and lead to misdiagnosis.

To solve this, researchers need to develop "dynamic" reconstruction algorithms. However, high-quality, open-source raw data of complex biological objects is surprisingly scarce. This paper fills that gap by providing a specialized dataset of a ladybug, offering enough complexity and "dynamic" variety to test how algorithms handle motion and sparse data (using fewer X-ray projections to reduce radiation dose).

Methods

The team used a Procon X-ray CT portable device to capture the data. The setup involved:

  • Hardware: An X-ray source (50 kV, 50 W) and a 1-megapixel detector with a 48 $\mu$m pixel size.
  • Geometry: A cone-beam setup with a Focus-to-Object Distance (FOD) of 111 mm and a Focus-to-Detector Distance (FDD) of 241 mm.
  • Acquisition: For each measurement, they captured 360 projections over a full 360-degree rotation.
  • Processing: The raw 2D projection images (1024 × 967 pixels) were processed into 3D sinograms. They specifically extracted 186 rows corresponding to the central horizontal cross-sections of the ladybug to form the final data files.

To simulate "dynamics," the researchers performed eight different measurements of the ladybug, each starting from a different position.

Datasets

The paper introduces a new, publicly available dataset hosted on Zenodo. It consists of:

  • Eight Sinogram Files: sinogram1.mat through sinogram8.mat.
  • Data Dimensions: Each file is a 3D array of $1024 \times 360 \times 186$ (Width $\times$ Projections $\times$ Rows).
  • Format: The data is provided as logarithmically transformed and normalized sinograms, ready for use in reconstruction frameworks like the ASTRA toolbox or opTomo for MATLAB.

Key Insights and Ideas

  • The Ladybug as a Proxy: While a ladybug isn't a human patient, its complex internal biological structure provides a much more rigorous test for algorithms than simple geometric "phantoms" (like plastic cylinders) often used in labs.
  • Simulating Motion: By providing eight distinct scans from different starting orientations, the authors allow researchers to "stitch" data together to simulate temporal changes or movement during a scan.
  • Sparse Data Testing: Because the dataset includes a dense 360 projections, researchers can intentionally "throw away" data (e.g., using only every 10th projection) to test how well new AI or iterative algorithms can reconstruct a clear image from very little information—a key goal for reducing medical radiation.

Key Results

The primary result is the successful documentation and release of the dataset. The authors provide:

  • Visual Verification: Initial 2D slice reconstructions using standard Filtered Backprojection (FBP) show clear internal morphology of the ladybug, confirming the data quality.
  • Mathematical Framework: The paper explicitly defines the measurement geometry and the linear model ($Ax = m$), making it easy for other researchers to build the necessary projection matrices ($A$) to test their own reconstruction solvers ($x$).

Limitations and Future Work

  • Discrete vs. Continuous Motion: The "dynamic" aspect is currently represented by eight static snapshots from different positions rather than a continuous, real-time video of a moving insect.
  • Software Dependency: While the data is open, the paper suggests using specific toolboxes (ASTRA, opTomo) which, while common in the field, may require a learning curve for newcomers.
  • Future Directions: The authors suggest this data is ripe for "limited data dynamic tomography," where algorithms must guess the object's shape when the scanning angle is restricted or the data is extremely sparse.

Summary updated

Abstract

This is the documentation of the 3D dynamic tomographic X-ray (CT) data of a ladybug. The open data set is available https://zenodo.org/record/3375488#.XV_T3vxS9oA and can be freely used for scientific purposes with appropriate references to the data and to this document in http://arxiv.org/
BibTeX
@misc{juurakko2019dynamic,
  author = {Juliaana Juurakko and Zenith Purisha and Simo Särkkä},
  title = {{Dynamic 3D Tomographic X-ray Data of Ladybug}},
  year = {2019},
  eprint = {1908.08782},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1908.08782},
}
Figure