Media 000669901: Knee

Restricted Download

Type Volumetric Image Series

Modality Synchrotron Imaging

Media represents biological specimen

am:4766 Heterodontosaurus tucki

Managed by

Jonah Choiniere

Synchrotron Imaging

European Synchrotron Radiation Facility (European Synchrotron Radiation Facility)

Fernandez, Vincent

--

--

Specimen AM 4766 was imaged on the ID17 beamline of The European Synchrotron (ESRF, Grenoble, France). The experiment consisted of two parts: (1) imaging the whole of each block and (2) re-imaging specific regions of interest at higher resolution (i.e., the sternal region where anatomy is obscured by matrix and larger parts of the skeleton). The beamline was setup for propagation phase-contrast synchrotron X-ray micro-computed tomography: a monochromatic beam of 160 keV (Laue double bent Si 111), a sample-detector distance of 11 m and an indirect detector comprising a 2 mm LuAG scintillator, a set of lenses with a magnification of ~0.3×, and a visible light detector (CCD or sCMOS). For the first set of experiments, we used a FReLoN 2k camera (ESRF), recording 4998 radiographs over a 360° rotation of the sample with a pixel size of 46.95 µm. Each radiograph consisted of five accumulated images, each with an exposure time of 40 ms, resulting in a total exposure time of 0.2 s per recorded radiograph. For the second part of the experiment, we used a PCO.edge 5.5 sCMOS with camera link (PCO, Kelheim, Germany), recorded 6000 projections with a pixel size of 21.80 µm. The total exposure time was 0.1 s (four accumulated images of 25 ms). For both sets of experiments, the centre of rotation was shifted near the edge of the field of view, almost doubling the size of reconstructed slices, keeping a margin ranging from 10% to 20% to ensure an overlap between projections 180° apart. Given the limited vertical size of the X-ray beam (ca. 7.5 mm), the full acquisition of a sample required several scans, moving various blocks of the specimen by 50% of the vertical field of view between each scan. The significant overlap was used to stitch radiographs vertically before tomographic reconstruction. As the PCO.edge 5.5 produces a vertically banding noise, each image was de-noised: first, a 1D high-pass median filter with a width of 5 pixels was applied horizontally; second, a three passes 1D vertical median filter with a width corresponding to 50% of the vertical size of the image. The overlapping part of two consecutive scans required a registration: the horizontal alignment was based on the position of the centre of rotation, while the vertical alignment was evaluated based on the mean error around the predicted value. Computation of the overlapping part was handled in two parts. The low frequencies (based on a low-pass median filter of a few pixels in width) were obtained using a weighted average, based on the vertical intensity profile of the beam. High frequencies (HFs) were handled to diminish the presence of strong ring artefacts in reconstructed slices: first, a blurred averaged of both HF images was generated; then a final image was generated, selecting pixels from both HF images that were less divergent from the blurred averaged image. Tomographic reconstruction was done using PyHST2 (Mirone et al., 2014) using the single distance phase retrieval approach (Paganin et al., 2002). The δ/β parameter (indicating the nature of the material for single-phase material) was set to reduce the noise while avoiding blurring of dense metallic inclusions. To reduce the impact of dense inclusions in the data, we performed a normalization of artifactual grey-level gradients and metallic inclusion correction (Cau et al., 2017). Remaining processing included modification of the bit depth from 32 to 16 bits, using the 0.001% minimum and maximum exclusion values of the 3D histogram generated by PyHST2; ring correction (Lyckegaard et al., 2011); and cropping the volume. Finally, a 2 × 2 × 2 binning was applied to increase the signal-noise-ratio and reduce the size of the data for segmentation.

None

None

--

--

None