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

Abstract

Recent research in computational imaging largely focuses on developing machine learning (ML) techniques for image reconstruction, which requires large-scale training datasets consisting of measurement data and ground-truth images. However, suitable experimental datasets for X-ray Computed Tomography (CT) are scarce, and methods are often developed and evaluated only on simulated data. We fill this gap by providing the community with a versatile, open 2D fan-beam CT dataset suitable for developing ML techniques for a range of image reconstruction tasks. To acquire it, we designed a sophisticated, semi-automatic scan procedure that utilizes a highly-flexible laboratory X-ray CT setup. A diverse mix of samples with high natural variability in shape and density was scanned slice-by-slice (5,000 slices in total) with high angular and spatial resolution and three different beam characteristics: A high-fidelity, a low-dose and a beam-hardening-inflicted mode. In addition, 750 out-of-distribution slices were scanned with sample and beam variations to accommodate robustness and segmentation tasks. We provide raw projection data, reference reconstructions and segmentations based on an open-source data processing pipeline.

Fig. 1

Overview of the scope of the 2DeteCT dataset.

Fig. 2

FleX-ray Lab: the computed tomography set-up used for the data acquisition. (1) cone-beam X-ray source; (2) Thoraeus filter sail; (3) Rotation stage; (4) Sample tube; (5) Flat panel detector. The objects 1, 3, 4, and 5 move from their red transparent front position to the mid position for the acquisitions of mode 3. In both positions 3,601 projection images per slice are taken while the object rotates 360 degrees.

Fig. 3

Beam spectra of a tungsten target X-ray source with an X-ray exit window made of 300 μm Beryllium operated at 60 kV with no added filtration and 90 kV tube voltage with no added filtration and filtered with a Thoraeus filter of Sn = 0.1 mm, Cu = 0.2 mm, Al = 0.5 mm simulated by TASMIP software.

Fig. 4

Visualization of the scanning procedure.

Fig. 5

From left to right: Sinograms and reconstructions from slice 1661 (mode1 -low-dose), slice 4300 (mode2 - artifact-free), slice 560 (mode3 - artifact-inflicted) and segmentation of slice 4300 based on the mode2 reconstruction. To illustrate the variation in the sample mix we selected different slices for each mode.