Converted BMD values, derived from routine clinical contrast-enhanced DLCT and adjusted for vessel iodine concentrations, showed a high agreement with non-enhanced DLCT-BMD. Moreover, the phase-independent conversion equation provides results which are adequate for the detection of low BMD in a clinical context as well.
Before this conversion, DLCT-BMD values were consistently higher for all scan phases when measured in contrast-enhanced scans compared with their native BMD references (Fig. 2). This represents the challenge of adequately separating intravascular iodine within the vertebra from HA, which is attributable to similar spectral absorption behavior of the two components. Another explanation for this BMD variation is the present dual-layer set-up, which cannot provide absolute selectivity on the detector level due to an overlap of the high- and low-energy spectra [25, 26].
As the very same VOIs were used for comparing both calculations, the utilization of spectral information is capable of additionally improving overall BMD accuracy. Although DLCT is still not perfectly specific for HA, the consistently higher coefficients of determination in the linear regression (Table 1) and the GEE analysis suggest that DLCT-BMD shows a more pronounced functional relation between contrast-enhanced and native scans compared with conventional BMD. Besides, quantifying iodine concentrations outside osseous structures, more precisely in large vessels (Vena portae, Aorta), significantly improves the accuracy of converted BMD.
In a pilot analysis (n = 12), tissue iodine concentrations of paraspinal muscle and fat were measured for all scan phases. Balancing the linear model between completeness and complexity, however, the rationale was to set up a slender, numerically stable model without overfitting or multicollinearity. Here, preliminary data suggested focusing on vessel iodine measurements in order to adjust for the influence of intravascular contrast agent. After the conversion, DLCT-BMD showed a high agreement between native and converted values in Bland-Altman plots (Fig. 3a–c) and a substantial reduction of MAEs (relative reduction of 64% for AR, 83% for PV, and 69% for PI). The validation with external data of the test cohort revealed coefficients of determination that are equivalent to those of the training cohort (Table 2). This finding confirms a numerically stable model for BMD prediction that is accurate in new data.
Particularly the phase-independent approach features critical advantages over phase-specific conversion equations as in clinical routine, the iodine concentration within the vertebra’s trabecular compartment may be affected by numerous factors apart from scan timing [27, 28]: e.g., contrast application speed, volume, the patient’s circulatory capacity, volume of distribution, red vs. yellow bone marrow ratio. BMD values derived from a phase-independent conversion equation could minimize the influence of those factors and therefore may be most useful in clinical reality.
Focusing on contrast-enhanced clinical DLCT exams, these routinely acquired clinical CT data were utilized for opportunistic BMD measurements to monitor changes such as tumor- or therapy-associated bone loss, when current non-enhanced scans are missing. The proposed DLCT-based method provides the opportunity to retrospectively screen for low BMD and could at once spare patients additional exposure to radiation by dedicated bone densitometric exams. Quantifying individual iodine concentrations in abdominal vessels turned out to be a practicable way of adjusting for the influence of intravascular contrast agent. In this context, there is extensive literature indicating the high accuracy of DLCT-based iodine quantification: within the typically encountered concentration ranges in clinical radiology, relative mean errors of about 3.3 to 4.6% were found, with simulated patient size and tube voltage inconsistently affecting measuring precision [20, 29, 30].
Comparable studies assessing iodine-associated effects on BMD quantification with multidetector CT (MDCT) encountered limitations: a study by Baum et al had a small training cohort for the conversion equation, relied on phantom calibration and only used PV scans . A comparable study by Kaesmacher et al was limited by a small number of enrolled patients .
The conversion equations in this DLCT study, however, are based on a solid training cohort of 88 patients. An additional correction step was introduced by adjusting for vascular iodine concentrations, which can be obtained with minimal effort in work and time. Besides, DLCT potentially combines the inherent benefits of dual-energy imaging for osteodensitometry with the major advantage of QCT, i.e., exclusive volumetric measurements of the trabecular compartment, which is more sensitive regarding therapy-associated bone remodeling processes —however, without needing synchronous phantom calibration.
Note that the present statistical models already show very high determination coefficients for conventional BMD: contrary to a comparable study investigating asynchronously calibrated BMD derived from contrast-enhanced MDCT, DLCT-BMD has substantially narrower 95% limits of agreement (− 10 to + 11 mg/ml (DLCT) vs. ca. − 30 to + 14 mg/ml (MDCT) for AR and − 10 to + 10 mg/ml (DLCT) vs. ca. − 39 to + 8 mg/ml for PV) and a better linear fit (R2: 0.983 vs. 0.923 for AR, 0.976 vs. 0.904 for PV) . Considering the minor contribution of the vessel iodine corrections, these results suggest a notably higher measuring accuracy of the DLCT scanner compared with MDCT.
Apart from several phantom studies, two in vivo trials already showed the diagnostic accuracy of native DLCT regarding osteodensitometric applications: Van Hedent et al demonstrated that DLCT-based BMD measurements perform very well in the detection of decreased BMD using DXA as standard of reference . A previously mentioned study by Roski et al showed that non-enhanced DLCT-based BMD measurements are on a par with phantom-based QCT .
This study has limitations. As CT exams were retrieved from clinical routine, there was no systematic variation in the amount of applied contrast agent to adjust for contrast load. Furthermore, neither overall circulatory parameters nor the local vascularization of the vertebral bodies for correlating contrast distribution could be investigated according to the retrospective nature of this analysis. Moderating both scan protocol inconsistencies and varying circulatory parameters, the phase-independent analysis is potentially meeting clinical reality best. Besides, the vertebral VOIs were placed manually, which contributes to the risk of a higher intra- or interobserver variability. A next step would be to overcome this issue by implementing a BMD analysis pipeline drawing on automatic segmentation. Additional longitudinal studies will be needed to investigate the in vivo reproducibility and the predictive power regarding incidental fractures.
In summary, this study showed that BMD values can be accurately estimated from contrast-enhanced multiphasic dual-layer spectral CT examinations, even independently from the used contrast phase. Moreover, measuring only one abdominal vessel for iodine concentration could significantly increase the goodness-of-fit in statistical models. Therefore, iodine-adjusted DLCT-BMD measurements suggest their potential value for a reliable opportunistic assessment of BMD even in routine clinical contrast-enhanced examinations.