I first derived a HU-density calibration using blocks of Porites skeleton with average thickness of approximately 1.2 cm. The relationship between HU and bulk skeletal density was:
HU = 519.27 × density - 833.92 (p <0.05; r2 = 0.96)
where density is in g cm-3 and the calibration was determined over the range of 1.04 to 1.20 g cm-3.
Next, I prepared two separate blocks of Porites skeleton (Fig. 1A), one with an average thickness of 1.2 cm (volume 4.04 cm3) and the other with an average thickness of 2.5 cm (volume 12.5 cm3), and I tested the effects of various scanning parameters on the derived densities (Fig. 1C-D). The density of the 1.2 cm block (calculated from measured mass and CT-derived volume) was 1.232 g cm-3. Based on the relationship above, this density should correspond to mean HU of -194 ± 7, which is within error of the measured mean HU of -195.2. The larger 2.5 cm block had a density (calculated from measured mass and CT-derived volume) of 1.029 g cm-3, corresponding to an expected HU of -300 ± 7. However, the measured mean HU was -337, outside the range of HU expected from the HU-density calibration. If the measured mean HU is applied to the calibration above, the calculated density is 0.956 g cm-3, 7% lower than the actual density. Thus, the thickness of skeleton in the scan appears to influence the measured HU, and by extension, the derived skeletal density. This demonstrates the importance of using skeletal standards of similar size to samples.
I tested the effects of several scanning and reconstruction parameters on the measured HU of the larger, 2.5 cm block. The initial scan of this block described above (with mean HU of -337), and the scans to establish the reported HU-density calibration, were all conducted at 89 kV and 272 μA, with 36 μm isotropic voxels, and reconstructed using a beam hardening correction of 19%. I then adjusted parameters of the sample scan without modifying the standard scans. Changing the sample scan to 80 kV and 306 μA resulted in a mean HU of -142, which corresponds to an apparent 39% increase in derived density. Increasing the beam hardening correction from 19% to 79% resulted in a mean HU of -118, or a 44% increase in derived density. Finally, decreasing the voxel size from 36 μm to 18 μm resulted in a mean HU of -312, or a 5% increase in derived density. This latter effect could result from partial-volume effects and/or fractal porosity. In materials with relatively large density contrasts (such as boundaries between air and skeleton in the pore holes of coral skeleton), the fraction of each voxel filled with the more dense material can influence the HU in a nonlinear manner. As finer spatial resolution includes more voxels along these boundaries, the partial-volume effect increases and/or smaller pore spaces are resolved if they follow fractal geometry (i.e. repeated patterns of pore spaces across spatial scales).
Overall, these results highlight the importance of scanning standards and samples with exactly the same settings because they have strong influences on the resulting HU. Note that these scans were conducted on the same day without moving the sample or turning off the X-ray source. In the case of the influence of beam hardening, the HU were derived from the same scan as the original while changing only the reconstruction settings. Further, the sample volumes calculated from the analysis program (see Methods) were all similar, less than 3% different between the initial scan and those in which the settings were changed. Thus, the changes in HU represent real effects of these parameters and are not artefacts of the analysis or sample positioning in the scan bed.
Finally, I tested whether the HU-density calibration determined using Porites standards could be accurately applied to a Pocillopora sample. Pocillopora has many small branches (Figure 1b) and thus it was not possible to cut the skeleton in a block as with the Porites samples, however the average thickness was approximately 1 cm. The density determined by mass and volume was 1.852 g cm-3. This is well outside of the density range of the Porites standards, but nevertheless the expected HU would be 128 ± 43. The measured mean HU was 147, which represents a 5% difference in density but is still within error due to the relatively large uncertainty associated with extrapolating the calibration to this high density.