Recent studies suggested that PCAT metabolic activity contributes to the development of coronary atherosclerosis [1, 2] and that FAI measurements reflecting PCAT attenuation in CT may represent a surrogate marker for coronary inflammation [6]. However, CT studies investigating this issue are subject to considerable variation in regard to CT scanner types, scan and contrast media protocols, and image reconstruction parameters [3, 4, 7,8,9,10, 13, 20,21,22,23, 30, 31].

Our study demonstrated an excellent intra- and inter-reader agreement for FAI measurements in presumably healthy subjects showing no evidence of CAD in CCTA. However, FAI values differed considerably both between subjects (up to 35 HU between individuals) and within subjects depending on both kernels and iterative reconstruction levels (each, p < 0.001), with noise changing across different reconstruction levels and indirectly influencing FAI as well. We found differences in FAI attenuation of, on average, 19 HU within the same subjects exclusively related to variable kernels and iterative reconstruction levels.

One of the major findings of this study was the relatively large inter-individual variations in FAI. Inflammation of the pericoronary adipose tissue in at least some of our subjects cannot be excluded, despite the fact that our subjects were otherwise healthy, had a low risk of CAD, no other comorbidities, and showed no CAD in CCTA examinations with excellent image quality. Importantly, these inter-individual variations were related to kernels and iterative reconstruction levels. The sharp vascular kernel (Bv56) without iterative reconstruction (QIR off) showed similar FAI values between subjects (average difference of only 10 HU), suggesting that pericoronary inflammation might not be the reason for these discrepancies in PCAT attenuation, but rather the type of image reconstructions.

Interestingly, the use of different kernels and iterative reconstruction levels led to an increase in this inter-individual difference in PCAT attenuation of up to 35 HU (quantitative smooth kernel Qr36 and QIR4). This difference between subjects substantially exceeds the PCAT attenuation differences of two recent metanalyses in the literature about pericoronary FAI, with reported differences of 6 HU [32] and 5 HU [5] between healthy subjects and patients with CAD. We strongly believe that these reported differences in PCAT attenuation related to reconstruction settings must be considered for follow-up CCTA studies investigating the topic of pericoronary inflammation.

In addition to inter-individual variations, we found considerable intra-individual variations in PCAT attenuation again related to kernels and iterative reconstruction levels. This may be explained, in the absence of evident clinical causes, through the CT value histograms and ESF curves. For the vascular kernels, the FAI decreased with increasing kernel sharpness at the same iterative reconstruction level. This effect is caused by the reduced range of edge smoothing with increasing kernel sharpness. As a result, the CT values measured in the PCAT near the edge of the vessel are less elevated, and the FAI decreases. For the smooth quantitative kernel, FAI increased with increasing iterative reconstruction levels despite of the QIR-independent ESF. This effect is a result of the reduced image noise at increasing QIR strength levels as shown by our results and as previously shown [28]. In the histogram plot, the peak of the CT values in the PCAT is significantly narrower at higher iterative reconstruction levels due to decreasing image noise. By limiting the FAI evaluation to the range of − 190 HU to − 30 HU (which is the standard for FAI assessment [3, 4]), only a part of this peak is asymmetrically averaged, and the FAI thus shifts to higher values as the peak becomes narrower. This effect is less pronounced for the vascular kernels because a complex interplay takes place between reduced image noise (with a tendency towards higher FAI values) and sharper edge display (with a tendency towards lower FAI values) as the iterative reconstruction levels increase.

Both reconstruction kernels and iterative reconstruction algorithms are manufacturer- and often also CT scanner-dependent. Previous PCAT studies included datasets with different reconstruction kernels (for example, a medium soft tissue convolution kernel Bf26 in the study by Dai et al [4], a different medium smooth kernel Bl26 in the study by Xi et al [8], and Bv36 in the study by Moser et al [9]), and the majority did not even report the kernels [3, 4, 7, 10, 20,21,22,23, 30, 31]. The same holds true for iterative reconstruction algorithms. Some studies reported the type and strength of iterative reconstruction (for example, Xi et al [8] and Moser et al [9] used sinogram-affirmed iterative reconstruction strength level 3), while others did not report on that issue [3, 4, 7, 10, 20,21,22,23, 30, 31]. Reconstruction-specific differences of PCAT may be further aggravated when using monoenergetic image reconstruction of spectral CT data with different keV levels [33]. Mergen et al [33] showed that the FAI of the RCA increased from − 89 ± 8 HU at 55 keV to −77 ± 12 HU at 80 keV. Our findings, despite being limited to a single PCD-CT scanner acquisition and reconstruction parameters, confirm previous evidence by Chen et al [34], who demonstrated with a conventional energy-integrating detector 256-row CT scanner, that PCAT attenuation linearly correlates with iterative reconstructions (r-squared > 0.99). These findings further underscore that reconstruction kernels and iterative reconstruction algorithms influence PCAT attenuation, both on new generation and conventional CT scanners, hereby not only expanding the generalizability of our results but also necessitating strong efforts in image acquisition and reconstruction standardization to allow for meaningful interpretations [35].

We used individualized contrast media injection protocols to compensate for inter-individual differences in patient size with the aim of achieving a relatively constant contrast attenuation in the coronary arteries. In our 20 patients, the median attenuation at the level of the origin of the RCA was 865, with an IQR of 770 to 904 at a monoenergetic level of 55 keV. We believe that such relatively constant vessel attenuation is important given the known effect of contrast attenuation on PCAT attenuation, as recently shown in the experimental porcine heart study by Pitteloud et al [36].

The following limitations of our study merit consideration. First, this was a single-center study with a small number of subjects. Second, both the perivascular analysis software and the CT scanner were limited to a single vendor, despite other software tools and scanners being available for these purposes as well. Third, the results were derived from a selected population of presumably healthy subjects, and no data were available for comparison to patients affected by acute and chronic disease. Finally, we cannot exclude pericoronary inflammation in our subjects although we tried to select apparently healthy participants in our study. However, we demonstrated in this study, among other results, large differences also within subjects, only by using different reconstruction settings.

In conclusion, our in-vivo study indicated that PCAT attenuation measured in CT heavily depends on kernels and iterative reconstructions. This further emphasizes the requirement for standardization of both CT data acquisition and reconstruction parameters to meaningfully demonstrate differences both within and across patient populations.