Radiologists may one day be able to skip an entire radiation-dosing step in many CT examinations, according to preliminary clinical results from a prototype deep silicon photon-counting detector CT scanner. In a small prospective pilot study, researchers compared virtual unenhanced images, synthetic images generated from contrast-enhanced scans, with true unenhanced images acquired in the same patients. Across the anatomic structures tested, 82 percent showed a difference of less than 10 Hounsfield units between the virtual and true unenhanced images, and 62 percent differed by less than 5 Hounsfield units. Because those thresholds are widely used benchmarks for judging whether synthetic unenhanced images are accurate enough for clinical work, the findings suggest that the prototype scanner and its companion image-processing algorithm come remarkably close to reproducing images that normally require a separate, unenhanced acquisition.
Virtual unenhanced images are not new. They have long been available on dual-energy CT systems, which acquire spectral information that allows software to subtract the signal of injected iodine contrast from an enhanced scan, leaving an image that approximates what the same tissue would look like without contrast. Radiologists use these synthetic images in two main ways: to provide a non-contrast phase when a protocol does not include one, such as in routine abdominopelvic imaging, and to replace a dedicated true unenhanced phase in protocols that routinely require it, such as those for mesenteric ischemia, active bleeding, and urography. Replacing the true unenhanced acquisition offers potential savings in both radiation dose and scan time, and it would eliminate any patient motion between the non-contrast and contrast-enhanced phases of a multiphasic examination.
Despite promising performance in applications such as characterizing enhancing renal masses with a single-phase protocol, virtual unenhanced imaging has been held back by variability between vendors, which has limited its widespread clinical utility. Photon-counting detector CT is an emerging technology that may change that picture. Compared with conventional CT, photon-counting systems offer increased spatial resolution, decreased noise, increased conspicuity of iodine, and the ability to perform spectral imaging prospectively without the hardware workarounds required by current dual-energy implementations, such as a special dual-layer detector, two separate X-ray tubes, or a system capable of rapidly switching tube currents. Current work also suggests that the spectral capabilities of photon-counting CT systems show improvement over dual-energy CT, although some studies have found dual-energy systems superior in certain situations, such as with large phantoms or dense materials.
The new study, conducted at the University of Wisconsin, tested a prototype deep silicon photon-counting CT system from GE Healthcare together with a prototype virtual unenhanced algorithm. Five adult subjects were scanned across six anatomic regions: one chest, one head, one wrist or hand, and three abdomens. Each subject was imaged in four phases: true unenhanced, portal venous, arterial, and late arterial, all at 120 kilovoltage peaks with tube current adjusted for anatomy and subject size and a slice thickness of 0.42 millimeters. Subjects received the iodinated contrast agent iohexol at 350 milligrams per milliliter, dosed by patient weight according to institutional protocols. The vendor reconstructed the images using the prototype algorithm, and the researchers then measured how closely the synthetic unenhanced images matched the true ones.
To quantify performance, the team selected five consecutive thin slices and three consecutive thick slices from each image set at two locations chosen to encompass vessels, soft tissue, and adipose tissue. Circular regions of interest were placed on the contrast-enhanced phases, made as large as possible over each structure, and then transferred to the corresponding locations on the true and virtual unenhanced images, with visual checks to account for patient motion between phases. The CT number of each structure was averaged over the slices, and the virtual unenhanced error was defined as the difference in Hounsfield units between the virtual and true unenhanced values. Published literature has used cutoffs of 5 and 10 Hounsfield units as metrics of virtual unenhanced performance, and the researchers adapted that methodology for this work.
The results varied by tissue type in an instructive way. For vessels, 70.4 percent of tested structures showed errors below 10 Hounsfield units, but only 40.7 percent fell below 5 Hounsfield units, and the average absolute error for vessels was 9.8 Hounsfield units, the highest of any category. Soft tissue performed best: 96.9 percent of structures were within 10 Hounsfield units and 84.4 percent within 5, with an average absolute error of just 2.6 Hounsfield units. Adipose tissue fell in between, with 75 percent within 10 Hounsfield units, 58.4 percent within 5, and an average absolute error of 4.9 Hounsfield units. Small vessels and adipose tissue were the structures most likely to exceed both error thresholds. The authors note that bones, which were not assessed in this study, accounted for much of the residual error visible on a subtraction-based error map, which also revealed some patient movement between contrast phases.
Clinical example images reinforced the quantitative findings. In representative cases, virtual unenhanced images generated from portal venous, arterial, and late arterial phases showed that subtraction of iodine signal produced CT numbers similar to those on the true unenhanced images, with the aorta, portal veins, inferior vena cava, and other vessels visibly well subtracted in several subjects. The study also included a control comparison in which the virtual unenhanced algorithm was applied to the true unenhanced images themselves, providing a baseline for judging the algorithm’s behavior. Taken together, the measurements and images indicate that the prototype deep silicon system can remove iodine signal from contrast-enhanced scans with small residual differences in attenuation across most tested tissues.
The findings align with a growing body of literature on photon-counting CT virtual unenhanced imaging. Previous work found virtual unenhanced errors ranging from 1 to 5 Hounsfield units depending on the structure analyzed and the contrast phase. One 2024 study of a commercially available cadmium telluride photon-counting system reported errors from less than 1 Hounsfield unit in the liver during the arterial phase up to 24 Hounsfield units in adipose tissue during the arterial phase, with errors exceeding 10 Hounsfield units for the aorta and adipose tissue in both arterial and venous phases; the present work, by contrast, showed errors below 10 Hounsfield units for both the aorta and adipose in arterial and venous phases. Another 2024 group reported errors below 1 Hounsfield unit in the arterial phase and below 3 in the venous phase for spleen, liver, and muscle, while the corresponding errors here were only slightly larger, at 2 Hounsfield units in the arterial phase and 4.5 in the venous phase. A 2022 study of 15 patients found a mean error of 3.7 Hounsfield units, similar to the soft tissue and adipose results reported here.
The authors argue that this consistency across scanner technologies preliminarily supports vendor-neutral performance of virtual unenhanced images in the clinical setting, an issue that could not be resolved with previous dual-energy implementations because technological differences between dual-energy systems affected the generalizability of results. If a deep silicon photon-counting system performs comparably to cadmium telluride-based photon-counting systems, the inter-vendor variability that has hindered clinical adoption of virtual unenhanced imaging may decrease. The researchers write that in a larger patient population they expect the deep silicon photon-counting CT to adequately subtract iodine signal from contrast-enhanced images to produce accurate virtual unenhanced images, and that future multiphasic CT protocols on this scanner may not require a dedicated true unenhanced phase, instead substituting the virtual version to reduce dose and scan time, as has been seen in the dual-energy CT literature.
Substantial caveats remain, and the authors are explicit about them. Only five subjects were enrolled, which limits the generalizability of the results and precludes meaningful statistical testing. Both the scanner and the virtual unenhanced algorithm are prototypes that could change before commercial release. The study excluded pregnant or lactating patients, patients allergic to iodinated contrast, those with chronic kidney disease, and those needing emergent care, and enrolled adults who had undergone imaging at the same hospital within the previous four months. Planned future work includes phantom-based error analysis, a large clinical dataset covering a wide range of acquired contrast-enhanced phases, and a reader study to strengthen and generalize the findings. Even so, for dose-sensitive populations such as pediatric, pregnant, or frequently scanned patients, the prospect of leveraging photon-counting CT’s advertised benefits in spatial resolution, noise, and iodine contrast while eliminating an entire acquisition phase makes these early results noteworthy for the field.
Subject of Research: Accuracy of virtual unenhanced images generated by a prototype deep silicon photon-counting detector CT scanner compared with true unenhanced images
Article Title: Performance of virtual unenhanced images on a prototype silicon photon counting detector CT: preliminary clinical results
Article References: Salyapongse, A. M., Szczykutowicz, T. P., Lubner, M. G., Yin, Z., Yan, M., Yue, M., McClure, K., & Toia, G. V. (2026). Performance of virtual unenhanced images on a prototype silicon photon counting detector CT: preliminary clinical results. BMC Biomedical Engineering, 8(1), Article 3. https://doi.org/10.1186/s42490-026-00103-0
Image Credits: AI Generated
DOI: 10.1186/s42490-026-00103-0
Keywords: photon-counting CT, virtual unenhanced imaging, dual-energy CT, Hounsfield units, iodine contrast subtraction, radiation dose reduction, deep silicon detector, computed tomography, spectral imaging, pilot study, contrast-enhanced CT, medical imaging
News Source: Ophelia Keating. (October 30, 2026). Prototype Silicon Photon-Counting CT Produces Virtual Unenhanced Images Close to True Scans. Scienmag.
