Quantification of periaortic adipose tissue in contrast-enhanced CT angiography: technical feasibility and methodological considerations

Figures

• 

  Fig. 1. a Summary of PaFT measurements. The top images show the reconstructed volumes measured (from left to right: aortic Volume, periaortic Volume, native PaFT Volume and arterial PaFT Volume. Note that the top views of the 3D reconstructed PaFT volumes show the inside of the cylinder containing different aortic contours and not fat-containing voxels within the aortic disc). The bottom images show the respective ROIs used for the reconstructed volumes (from left to right: native aortic volume, native and aortic periaortic volumes, native PaFT volume and arterial PaFT volume). b Distribution of pixel HU values within the periaortic ROI in the native and arterial scans (the red line represents the upper threshold for the fat tissue containing pixels with values − 45 HU to − 195 HU). In the native phase pixel HU values peak at about 40 HU. In the arterial phase pixel values within the identical periaortic ROI peak at about 356 HU. The distribution of pixels with values below − 45 HU shows few changes. c Distribution of pixel HU values within the PaFT-ROI in the native and arterial scans (the blue line indicates the range of HU values containing the peak HU value). The distribution of HU…

• 

  Fig. 2. Univariate linear regression in the derivation study. For PaFT Volume including (a) and excluding (b) large AAAs and for the mean PaFT HU value including (c) and excluding (d) large AAAs the slope of the line of best fit was the correction factor for conversion of arterial to native PaFT values

• 

  Fig. 3. a, b Comparison of corrected arterial PaFT Volume and native PaFT Volume. The Bland–Altman plot on the left [mean difference 0.359 (− 0.0141 to 0.732), limits if agreement − 2.046 (− 2.688 to − 1.404) and 2.764 (2.121–3.406)] after removal of three outliers shows a good dispersion of values with only one value outside the LoA. The blue line of zero difference is inside the bar representing the 95% CI of the mean difference. The Passing–Bablok plot on the right showed only minimal systematic bias. The Cusum test (p = 0.59) and Spearman coefficient (r = 0.983, p < 0.0001) showed no significant deviation from linearity. c, d Comparison of corrected arterial PaFT mean HU value and native PaFT mean Hu value. The Bland–Altman plot on the left [mean difference − 0.835 (− 1.775–0.106), limits if agreement − 6.823 (− 8.443 to − 5.204) and 5.154 (3.534–6.773)] after removal of three outliers shows a good dispersion of values with only one value outside the LoA. The blue line of zero difference is inside the bar indicating the CI of the mean difference. The Passing–Bablok plot on the right showed no significant proportional or systemic bias. The Cusum test (p = 0.56) and Spearman…

• 

  Fig. 4. Secondary study results. Box–Whisker plots (medians) for PaFT “ratios” [PaFT volumes] in the non-AAA vs. AAA groups measured in native (0.462 vs. 0.513) (a) and corrected arterial (0.449 vs. 0.515) (b) scans, as well as for mean PaFT HU values in the native (− 77.7 vs. − 80.0) (c) and corrected arterial (− 77.0 vs. − 77.8) (d) scans. The Mann–Whitney test showed no significant difference between the non-AA and AAA groups

• 

  Fig. 5. Methodological considerations/limitations. Presence of intraluminal pixels with values − 45 to − 195 HU in native CT scans. Usually, only solitary pixels in few axial CT images are seen (a). Such pixels appear in the vicinity of metal/high density foreign objects, for instance near spinal osteosynthesis materials (b, c). Alternatively, they can arise near very high-density calcium deposits in the aorta or vertebral column (d). Distribution of unenhanced CT scans of the derivation cohort with increasing number of intraluminal voxels with values of − 45 to − 195 HU (e). More than half of the CT-scans (55/101) had zero voxels, two-thirds (67/101) had 1 or less voxels, 83/101 had 2 or less voxels and 91/101 had 10 or less voxels. Of the 10 examinations with > 10 voxels the number of false positive voxels was still miniscule compared to the total number of fat-containing voxels (typically 1500–25,000). Considering the difference of fat-containing voxels between the unenhanced and enhanced data sets, the false positive voxels of these ten last examinations would account only for 0.27–2.67% of the measured difference in voxels − 45 HU to − 195 HU between the unenhanced and enhanced…