Coronary and vascular calcifications influence how radiologists detect, characterize, and quantify disease, and their visibility depends not only on how big or dense a lesion is but on how the image was made. The headline result is a trade-off rather than a winner: 68-keV imaging delivered the highest contrast-to-noise ratio for calcium-mimicking inserts, but it also carried the highest radiation dose, and once dose was factored in, the three strategies looked far more alike.

The study, led by Yeu-Sheng Tyan, Ya-Yun Hsiao, Cheng-Chang Lu, and Tou-Rong Chen, was conducted entirely in phantoms, so no patients were involved and ethics approval was not required. The team used a QRM thorax phantom with external dimensions of roughly 200 by 300 millimeters and a height of 100 millimeters, fitted with a QRM cardiac calcification insert in its central bore. That insert held nine calcium-mimicking cylinders built from hydroxyapatite at three densities, 800, 400, and 200 milligrams per cubic centimeter, in three nominal sizes of 5, 3, and 1 millimeters in diameter and matching lengths. To simulate a larger body habitus, the researchers added a fat-equivalent extension ring around the phantom, creating a large-phantom condition with a greater effective cross-sectional area.

Scanning was performed on a GE Revolution Apex system using three acquisition strategies: routine-dose 120-kVp CT, low-dose 120-kVp CT, and dual-energy CT with rapid switching between 80 and 140 kVp, from which 68-keV monoenergetic images were reconstructed. The 68-keV level was chosen because it is a predefined setting in the authors’ workflow and is frequently discussed in relation to standard 120-kVp imaging. Every image set was reconstructed at a 2.5-millimeter slice thickness using a standard kernel and three reconstruction methods: filtered back projection, ASiR-V iterative reconstruction at 50 percent, and TrueFidelity deep-learning reconstruction at medium strength. Because filtered back projection, iterative, and deep-learning images came from the same raw data, the volume CT dose index was identical across reconstruction methods within each protocol and phantom condition.

Quantitative analysis was deliberately simple and transparent. For each protocol, phantom condition, and reconstruction method, two slices showing the highest insert conspicuity were selected and referenced by their original DICOM filenames. Nine circular regions of interest were placed at the centers of the calcium-mimicking inserts, plus one large background region in a homogeneous area, with exclusion circles keeping calcification contamination out of the background measurement. A Python workflow read the DICOM data directly, converted pixel values to Hounsfield units using the rescale slope and intercept, and extracted mean attenuation, background noise defined as the standard deviation of the background region, and the contrast-to-noise ratio, calculated as the absolute attenuation difference between insert and background divided by background noise. A figure of merit, defined as the contrast-to-noise ratio squared divided by the volume CT dose index, captured dose-normalized performance.

The primary analysis focused on the 5-millimeter and 3-millimeter inserts, which were consistently visualized at 2.5-millimeter slice thickness; the 1-millimeter inserts were treated as exploratory because their tiny size invites severe partial-volume averaging. Across the 5-millimeter and 3-millimeter targets, attenuation rose with hydroxyapatite density, a consistency check that reassured the authors that their region placements were sound. The with-ring, larger-phantom condition showed higher background noise and lower contrast-to-noise ratio in nearly every comparison, confirming the expected penalty of increased attenuation and scatter. Reconstruction method mattered just as consistently: deep-learning reconstruction generally produced the lowest noise and the highest contrast-to-noise ratio, iterative reconstruction came next, and filtered back projection showed the highest noise and the weakest conspicuity.

In the without-ring condition at 800 milligrams of hydroxyapatite per cubic centimeter, contrast-to-noise ratio for 5-millimeter inserts climbed from roughly 34.0 with routine-dose filtered back projection to 61.3 with 68-keV filtered back projection, and from 63.9 with routine-dose deep-learning reconstruction to 95.7 with 68-keV deep-learning reconstruction. For 3-millimeter inserts under the same conditions, the corresponding figures rose from about 19.7 to 33.8 with filtered back projection and from 36.7 to 52.1 with deep-learning reconstruction. In both phantom sizes and both insert sizes, the ranking of acquisition strategies held: 68-keV monoenergetic imaging first, routine-dose 120-kVp second, and low-dose 120-kVp last.

The volume CT dose index followed the same order as the contrast-to-noise ratio, highest for the 68-keV dual-energy protocol, intermediate for routine-dose 120-kVp, and lowest for low-dose 120-kVp. The authors note that the dual-energy protocol had the lowest noise index while the low-dose protocol had the highest, so the conspicuity gap cannot be attributed to energy alone; it partly reflects protocol-dependent noise targets and dose settings. When performance was normalized using the figure of merit, the differences among the three strategies shrank considerably. In the authors’ framing, the three acquisitions form a dose-performance continuum: 68-keV imaging buys conspicuity at a dose cost, low-dose 120-kVp saves dose at the cost of conspicuity, and routine-dose 120-kVp sits between them on both axes.

The study’s limitations are stated plainly and matter for interpretation. It was conducted on a single CT platform with a single set of reconstruction parameters, so results may be vendor- and system-specific. The analysis was descriptive, with no repeated scans and no repeated region-of-interest redraws, and no formal statistical testing, so the rankings should be read as condition-level comparisons rather than proven superiority. Only one monoenergetic level, 68 keV, was evaluated in detail, the protocols were not dose-matched, and the exclusive use of 2.5-millimeter reconstructions likely amplified partial-volume effects for smaller inserts. The 1-millimeter inserts illustrated that point vividly: their attenuation values often failed to follow the expected monotonic ranking with density, particularly in the larger phantom, and the authors concluded that reliable quantification of such targets would likely require thinner reconstructions or a dedicated high-resolution protocol.

The authors argue that the common assumption that 68-keV monoenergetic images simply approximate conventional 120-kVp CT is incomplete. In their data, 68-keV imaging was not merely similar to routine-dose 120-kVp CT; it generally produced higher contrast-to-noise ratios for both 5-millimeter and 3-millimeter calcification mimics. The more useful question, they suggest, is not whether the images look alike but how lesion conspicuity, radiation dose, and dose-normalized performance should be balanced when optimizing a calcium-focused CT protocol.

Because this was a technical phantom study on one scanner, the authors are careful to say that clinical validation remains necessary before the findings can guide routine calcium-focused CT tasks or patient-specific protocol selection. They call for further work across additional monoenergetic energy levels, CT vendors, reconstruction platforms, and patient-based datasets to confirm generalizability.

Subject of Research: A phantom study comparing 68-keV monoenergetic, routine-dose, and low-dose 120-kVp CT for calcium-mimicking inserts across phantom sizes and reconstruction methods.

Article Title: Quantitative comparison of 68-keV monoenergetic, routine-dose, and low-dose 120-kVp CT for calcium-mimicking inserts in thoracic phantoms: effects of phantom size and reconstruction method

Article References: Tyan, Y.-S., Hsiao, Y.-Y., Lu, C.-C., & Chen, T.-R. (2026). Quantitative comparison of 68-keV monoenergetic, routine-dose, and low-dose 120-kVp CT for calcium-mimicking inserts in thoracic phantoms: effects of phantom size and reconstruction method. European Radiology Experimental, 10(1), Article 145. https://doi.org/10.1186/s41747-026-00818-8

Image Credits: AI Generated

DOI: 10.1186/s41747-026-00818-8

Keywords: computed tomography, dual-energy CT, monoenergetic imaging, calcium, contrast-to-noise ratio, radiation dose, deep-learning reconstruction, phantom study, image quality, protocol optimization, Quantitative, comparison