Dataset Open Access
This submission contains a collection of 131 CT scans of pieces of modeling clay (Play-Doh) with various numbers of stones inserted. The submission is intended as raw supplementary material to reproduce the CT reconstructions and subsequent results in the paper titled "A tomographic workflow enabling deep learning for X-ray based foreign object detection" [Zeegers 2022]. This submission consists of three parts in total.
The 131 CT scans are divided into 5 separate submissions:
Part 1 of 5: 001-028: 10.5281/zenodo.5866228
Part 2 of 5: 029-056: 10.5281/zenodo.5866322 (this upload)
Part 3 of 5: 057-084: 10.5281/zenodo.5866363
Part 4 of 5: 085-111: 10.5281/zenodo.5866365
Part 5 of 5: 112-131: 10.5281/zenodo.5866367
The samples are modeling clay (Play-Doh, Hasbro, RI, USA) with various numbers of pieces of gravel included. In total 131 samples are prepared, of which 20 samples contain 5-8 inserted stones, 3 samples contain three stones, 35 contain two stones, 62 contain one stone and 11 contain no stones. The stones have an average diameter of ca. 7mm (ranging from 3mm to 11mm). The Play-Doh is remolded for every sample.
The dataset is acquired in the FleX-ray Laboratory, developed by TESCAN-XRE, located at CWI in Amsterdam. The CT scanner consists consists of a cone-beam microfocus polychromatic X-ray point source, and a 1944x1536 pixel, 14-bit, flat detector panel (Dexela1512NDT). Full details can be found in [Coban 2020].
For each sample, 1800 radiographs are collected by rotating the sample over 360 degrees in a circular and continuous motion. A peak voltage of 90kV is used, and the target power is set to 20W. The distance between the source and detector is 69.80 cm and the distance between the source and the object is 44.14 cm. An exposure time of 20 ms is used for each projection.
This data is the result of a demonstration of a workflow to collect annotated data for supervised machine learning for X-ray based object detection. The ground truth locations are retrieved by tomographic reconstruction, segmentation and virtual projections with the same acquisition angles. A detailed description for the workflow to obtain a training dataset is given in [Zeegers 2022].
All projections are unprocessed files, except that a binning been applied FleX-ray lab software. The resulting image sizes are 956x760. Flatfield images (averaged over 10 pre and 10 post radiographs) and darkfield images (averaged over 10 pre and 10 post images) are included with each object. All images are stored in .tif format. The data for samples with 0-3 stones are contained in parts 1 to 4, while the samples with 5-8 stones constitute part 5. The size of the completely unpacked dataset (all 5 parts) is ca. 343.5 GB.
The processed data (with generated ground truth) is made available in another (smaller) submission for object detection purposes: https://zenodo.org/record/5681008
These datasets are produced by the Computational Imaging group at Centrum Wiskunde & Informatica (CI-CWI) in Amsterdam, The Netherlands: https://www.cwi.nl/research/groups/computational-imaging
zeegers [at] cwi [dot] nl
The authors would like to acknowledge the funding from the Netherlands Organisation for Scientific Research (NWO), project number 639.073.506. The authors also acknowledge TESCAN-XRE NV for their collaboration and support of the FleX-ray laboratory.
[Zeegers 2022] M. T. Zeegers, T. van Leeuwen, D. M. Pelt, S. B. Coban, R. van Liere, K. J. Batenburg, "A tomographic workflow to enable deep learning for X-ray based foreign object detection", 2022 (submitted)
[Coban 2020] S. B. Coban, F. Lucka, W. J. Palenstijn, D. Van Loo, and K. J. Batenburg, “Explorative imaging and its implementation at the FleX-ray Laboratory,” J. Imaging, vol. 6, no. 18, 2020, doi: 10.3390/jimaging6040018.
If you use (parts of) this data in a publication, we would appreciate it if you would refer to the first article.