Calcium scoring using virtual non-contrast images from a dual-layer spectral detector CT: comparison to true non-contrast data and evaluation of proportionality factor in a large patient collective

Ischemic heart disease is the most common global cause of death, with high mortality reported in the developed countries [1]. Therefore, early detection of coronary artery disease (CAD) is crucial for the prevention of adverse events through dedicated therapy. Hence, non-invasive cardiac imaging by computed tomography plays an increasingly significant role in diagnosis of CAD [2‐4]. Besides contrast-enhanced imaging for the evaluation of vascular stenosis, several studies showed that the determination of coronary artery calcium scoring (CACS) is an essential prognostic factor and a strong and independent predictor of cardiovascular events, such as myocardial infarction and sudden cardiac death [3, 5, 6]. The most commonly used method to evaluate patients’ burden of CAD is the Agatston score, which measures the amount of calcium present in each lesion scaled by an attenuation factor and summed over all lesions [7].

Whereas currently both contrast-enhanced CT imaging for stenosis determination and non-contrast images for CACS are widely established for the diagnosis of CAD, several methods using spectral imaging including the possibility to generate virtual non-contrast (VNC) images for calculation of CACS have previously been proposed in multiple studies [8‐10]. Dual-energy imaging is based on simultaneous acquisition of two CT datasets at different x-ray spectra—either by data acquisition at two different x-ray tube voltages (dual-source CT, kV switching) or by energy separation in the detector (dual-layer CT). Based on decomposition into two base materials (soft tissue and iodine), a virtual non-contrast image can be generated from a contrast-enhanced image [11, 12]. Especially the novel technique of spectral CT imaging using a dual-layer spectral detector CT (SDCT) system might help to overcome some limitations of other dual-energy techniques. For example, there is no temporal or projection offset using the dual-layer method. Furthermore, the dual-layer method has been reported to be a reliable method of dual-energy imaging [13, 14].

In a preliminary study, Nadjiri et al showed that the use of spectral imaging with SDCT for the determination of CACS from contrast-enhanced coronary computed tomography angiography (coronary CTA) may be a feasible alternative and reaches good agreement with the conventional technique. Nevertheless, the results of CACS (Agatston score) from VNC images needed to be multiplied by a proportionality factor of 1.83 to match the results from TNC images due to underestimation of plaque density and plaque volume with the VNC data. Furthermore, the sample size was small and patients without coronary calcifications were included in statistical analysis, resulting in a higher correlation [5].

The use of VNC images generated from spectral data could reduce radiation exposure by omitting additional native scans when performing a CTA. For establishing this method in everyday clinical practice, we sought to evaluate a more representative patient cohort and compare VNC with true non-contrast imaging. Furthermore, this study analyzes the extent and causes of needed proportionality factors.

The determination of the coronary artery calcium score is an essential prognostic factor in patients with symptoms of coronary artery disease. Therefore, it is common practice to perform non-contrast-enhanced, electrocardiogram-gated scans for calcium quantification prior to CT angiography [17, 18]. Finding an approach for the determination of calcium score from contrast-enhanced images and thus omitting native scans preceding CT angiography would be useful for radiation dose reduction and shortening of the duration of the overall exam.

In this study including 103 patients with a correlation coefficient of 0.95, we showed that there is a very high correlation between results of calcium scoring from real non-contrast images as currently performed in clinical routine and VNC images derived from contrast-enhanced imaging of a 64-slice single-source dual-layer spectral CT system through a software package certified for medical use and commercially available.

These results in general are in line with previous studies which also have shown good results of calculation of calcium score from VNC images. Schwarz et al demonstrated a correlation coefficient of 0.95 in 36 patients in a dual-source CT system and Fuchs et al showed a high agreement between methods with a correlation coefficient of 0.96 in 52 patients using a rapid kVp-switching system and applying a reduced dose of contrast media [8, 9]. Nevertheless, this is the first study to evaluate the accuracy of CACS from VNC imaging computed from spectral data on a SDCT scanner in comparison to standard non-contrast imaging in a more representative patient cohort with clinically approved software.

Additionally, it is of note that contrary to other studies, patients with a CACS of zero were excluded in our study to better understand the correlation of the measured values [8, 9]. However, the intercept was 3.8 and therefore close to zero, indicating the correct association between the methods. Furthermore, of the 31 patients with CACS of zero, none exhibited virtual CACS of ≥ 5 indicating a low rate of false-positive results. The false-positive result was visually assessed and seems to be most likely caused by reconstruction artefacts as in very few occasions wall adherent contrast agent was not fully subtracted from the image.

Despite the exclusion of patients with a calcium score of zero, Pearson’s correlation coefficient of 0.95 still indicates one of the highest correlations of calcium scoring values between true and virtual non-contrast images and 83.3% of patients were classified in the same category comparing CACS calculated from TNC and VNC [8, 9, 19, 20].

Nevertheless, there is still a small residual discrepancy with 16.7% of patients being classified in a higher or lower risk category when using VNC as compared to TNC. This discrepancy does not necessarily need to be attributed to the difference in methods as a previous study showed a notable inter-scan variability within 5 min using the same CT equipment [21]. Additionally, different standard kernels used and processing through dedicated software might further influence the outcome of CACS [22].

For proper transfer of results between methods, a proportionality factor needs to be applied as analysis of real non-contrast images showed a calcium score 3.83 times as high as calcium scores analyzed in VNC images. As virtual non-contrast is based on decomposition into two different materials (soft tissue and iodine), attenuation of calcium can be reduced as a consequence. In a study published recently, Nadjiri et al showed that there might be two reasons for the underestimation of calcium scoring in VNC data: a slight underestimation of the plaque volume and, more importantly, an underestimation of plaque attenuation in VNC images; a reduced attenuation of Ca-plaques is to expect in the VNC reconstruction. Therefore, certain plaques will be excluded from the semiautomatic analysis as they will not exceed the threshold of 130 HU in the VNC reconstruction [5].

The goal of this study was to show a way of reducing radiation in analysis of CACS using a dual-layer SDCT. Tube-based dual-energy CT systems have been reported to increase radiation exposure depending on the patient’s heart rate, protocol, and scanner generation [23, 24]. Nevertheless, improved conventional detectors with higher dose efficiency as well as the omission of native scans reduced overall radiation exposure compared to current clinical standard procedures in CT-based evaluation of coronary artery disease by approximately 20–25% [25, 26]. When omitting the native scan in this study, effective dose could be reduced by 19.3%. The effective dose of 4.81 mSv for the contrast-enhanced scan with our scanner is comparable to radiation doses commonly published in the literature for this type of examination, although lower effective doses can be achieved when using e.g. 320-detector-row single-source CT scanners or dual-source CT scanners [27, 28].

Despite its clinical setting, high number of subjects, and good correlation, our study has several limitations. First of all, CT-based true non-contrast imaging was used as a reference standard. Although beam hardening and blooming have been described in the context of calcified coronary artery lesions and could lead to artifacts, they could also occur in native scans [8, 29]. Nevertheless, there is no other feasible method available in a clinical setting.

There are some differences in our results from this study compared to previously published findings, where Nadjiri et al reported a proportionality factor of 1.83. Some of these differences could be attributed to difference in sample size, exclusion of patients with CACS of zero, and different software and conversion tool used. Additionally, Nadjiri et al used a threshold of 90 HU for CACS whereas, in this study, the threshold was set to 130 HU as commonly used for Agatston scoring [5, 7]. For transfer into daily clinical use, confirmation of results in a multi-center study and precise definition of extent and causes of proportionality factor is needed.

To conclude, this is the first study to demonstrate that the determination of coronary artery calcium score is feasible using VNC images obtained from a dual-layer spectral detector CT in a representative patient cohort. By applying a correction factor, the results show good agreement with the standard technique. This could obviate the need for native scans typically used for CACS, thereby facilitating radiation dose reductions. Nevertheless, translation into clinical practice is subject to further studies and evaluation of causes of needed proportionality factor.