The research, led by Chuluunbaatar Otgonbaatar of the Medical Imaging AI Research Center at Canon Medical Systems Korea and colleagues at Yonsei University Wonju College of Medicine and the Mongolian National University of Medical Sciences, took advantage of a uniquely controlled setup. The team enrolled 202 adults, with a mean age of 65.5 years and 112 men among them, who underwent abdominopelvic CT between April and December 2025. Critically, every scan in the study was reconstructed from the exact same raw projection data using three different algorithms: filtered back projection, the classic workhorse of CT; hybrid iterative reconstruction, which refines images through repeated correction cycles; and deep learning reconstruction, the newest entrant that uses neural networks to suppress noise while preserving anatomical detail. Because the underlying x-ray data were identical, any differences in the final measurements could be attributed squarely to the reconstruction method rather than to variations in radiation dose, patient positioning, or biology.

To read out body composition values, the researchers deployed a fully automated deep learning segmentation tool, removing the human variability that has traditionally plagued quantitative imaging. The software measured adipose tissue density and area, muscle density and area, organ attenuation and volume at the third lumbar vertebral level, a standard anatomical landmark for body composition analysis, and calculated bone mineral density at the first and second lumbar vertebral bodies. Agreement across the three reconstruction algorithms was then evaluated using Bland-Altman analysis, which quantifies systematic bias and the 95 percent limits of agreement, together with intraclass correlation coefficients, a statistical measure of consistency that approaches 1.0 when measurements are essentially interchangeable.

The headline result is that the measurements clinicians rely on most heavily proved remarkably robust. Subcutaneous fat area, the fat layer just beneath the skin, showed a bias of just 0.05 square centimeters between filtered back projection and deep learning reconstruction, with 95 percent limits of agreement spanning a razor-thin range from minus 1.51 to 1.61 square centimeters. Skeletal muscle area, arguably the single most important biomarker in CT body composition analysis, differed by only minus 0.32 square centimeters between filtered back projection and hybrid iterative reconstruction, with limits of agreement from minus 1.42 to 0.76 square centimeters. Visceral fat area, the metabolically dangerous fat wrapped around internal organs, showed a bias of minus 0.56 square centimeters with limits from minus 2.04 to 0.91 square centimeters. Bone mineral density measurements were similarly stable, differing by just 0.12 milligrams per cubic centimeter between hybrid iterative and deep learning reconstruction. Across the evaluated measurements, intraclass correlation coefficients ranged from 0.953 to a perfect 1.000.

But the study also uncovered a genuine weak spot. While raw areas of muscle and fat barely budged across algorithms, the parameters that depend on attenuation thresholds proved far more sensitive. These measurements classify muscle tissue into low-attenuation and normal-attenuation categories based on Hounsfield unit cutoffs, a technique used to quantify fatty infiltration of muscle, itself a powerful marker of poor prognosis in cancer, liver disease, and aging. Because reconstruction algorithms alter the noise texture and local attenuation values of an image, pixels hovering near a classification threshold can flip categories, causing low-attenuation and normal-attenuation muscle areas to vary substantially between reconstructions even though the total muscle area stays constant. The lowest intraclass correlation coefficient in the study, 0.953, was observed precisely for low-attenuation muscle area between filtered back projection and deep learning reconstruction, and most area-based measures other than these threshold-dependent ones showed small absolute differences of only one to two square centimeters.

The physics behind this vulnerability is worth unpacking. Filtered back projection reconstructs images by mathematically inverting the projection process, a fast and elegant method but one that leaves images relatively noisy. Iterative reconstruction repeatedly compares simulated projections with the measured data and refines the image, trading computation time for noise reduction. Deep learning reconstruction goes further still, training neural networks on vast image datasets to distinguish noise from true anatomy, achieving what the study confirmed with noise power spectrum analysis: a progressive reduction in image noise moving from filtered back projection through hybrid iterative reconstruction to deep learning reconstruction. That noise reduction is wonderful for diagnostic image quality, but because computed attenuation values shift subtly as noise characteristics change, any measurement that draws a hard line at a specific attenuation value inherits that instability.

The implications ripple outward across modern medicine. CT body composition analysis is exploding in clinical relevance, guiding oncology dosing, predicting outcomes in surgery and transplantation, characterizing sarcopenia and obesity, and feeding large epidemiological studies that mine millions of existing scans. Automated AI segmentation tools are multiplying rapidly, promising to make these analyses routine rather than labor-intensive. But if a hospital upgrades its scanners and switches from iterative to deep learning reconstruction, could longitudinal measurements of a patient’s muscle quality appear to change even when the patient’s actual muscle has not? The new findings suggest that for simple area and volume measurements, the answer is reassuringly no, at least within a single scanner and protocol, but for attenuation-threshold biomarkers such as low-attenuation muscle area, reconstruction settings genuinely matter and could confound trend analyses.

The study’s authors are careful to frame their conclusions within the limits of their design. The work was conducted under a single-scanner protocol at a single institution, approved by the Institutional Review Board of Wonju Severance Christian Hospital with the requirement for informed consent waived given its retrospective nature. The findings, they emphasize, support the use of consistent reconstruction settings whenever attenuation-dependent biomarkers are involved, and they call for validation across different scanners, scanning protocols, software platforms, and longitudinal examinations before the results can be generalized. Two of the co-authors, Otgonbaatar Chuluunbaatar and Hackjoon Shim, are employees of Canon Medical Systems Korea, a subsidiary of Canon Medical Systems Corporation, a competing interest the journal discloses, though the other authors report no relevant financial interests.

Still, the study delivers a message that will shape how quantitative CT is deployed in the AI era. The marriage of deep learning reconstruction and deep learning segmentation is inevitable, offering cleaner images and fully automated biomarker extraction from scans already acquired for other reasons. This research provides the kind of rigorous, physics-aware validation that must accompany that marriage, demonstrating that the raw anatomical quantities survive the reconstruction pipeline essentially intact, while sounding a precise and actionable warning about derived, threshold-based measures. As hospitals worldwide modernize their CT fleets, the practical prescription is clear: lock in your reconstruction settings, document them, and treat attenuation-threshold biomarkers as sensitive to the imaging pipeline itself. In quantitative imaging, the algorithm behind the picture can matter almost as much as the picture itself.

Subject of Research: Effect of CT image reconstruction algorithms on automated body composition and bone mineral density measurements

Article Title: Agreement of automated computed tomography body composition measurement across image reconstruction algorithms under identical projection data

Article References: Otgonbaatar, C., Kim, H., Kim, J. W., Dagvasumberel, G., Cha, S.-J., Jeon, S.-H., Shim, H., & Ahn, J.-H. (2026). Agreement of automated computed tomography body composition measurement across image reconstruction algorithms under identical projection data. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02938-4

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02938-4

Keywords: computed tomography, body composition, deep learning reconstruction, iterative reconstruction, filtered back projection, skeletal muscle area, adipose tissue, bone mineral density, Bland-Altman analysis, medical imaging AI, sarcopenia, radiology

The research, led by Xing Zhang and Yong Peng of the Wuhan Hospital of Traditional Chinese and Western Medicine together with their colleagues, took advantage of an elegant study design known as a within-subject evaluation. Instead of comparing different patients scanned under different conditions—a comparison inevitably muddied by differences in body habitus, lung density, and underlying disease—the team acquired raw CT data from 36 patients once and then reconstructed the very same data multiple times. Each patient’s scan was processed with conventional filtered back projection, the workhorse algorithm of CT for decades, and then again with KARL 3D iterative reconstruction at three escalating strength levels: 3, 6, and 9. Because every reconstruction derived from identical raw data, any difference in image quality could be attributed cleanly to the algorithm itself rather than to biology or chance.

The technical setup matters for understanding what the team measured. All scans were performed at 120 kilovoltage peak with a mean volume CT dose index of 12.0 milligrays, a routine chest protocol. The lung window is notoriously the harshest test for any reconstruction algorithm, because the sharp, edge-enhancing kernels used to visualize lung parenchyma inherently amplify noise. The researchers placed a small circular region of interest, roughly 20 square millimeters, in the lung parenchyma just behind the right main bronchus at the level of the carina, deliberately avoiding vessels and airways. Within that tiny patch of tissue, they measured the standard deviation of pixel values in Hounsfield units—the standard, if imperfect, proxy for image noise—and calculated the signal-to-noise ratio.

The numbers tell a dramatic story. With filtered back projection, lung parenchyma noise averaged 51.76 Hounsfield units, with a standard deviation across patients of 13.31. At KARL 3D level 9, the strongest setting tested, that figure fell to 33.71 Hounsfield units—a reduction of nearly 35 percent. Signal-to-noise ratio climbed in lockstep, from 18.17 to 27.61. Crucially, every pairwise comparison between reconstruction settings reached statistical significance, and the repeated-measures analysis of variance produced an F statistic of 338.7 for the noise measurements, an extraordinarily large value that reflects how consistently the algorithm’s effect manifested across all 36 subjects. The denoising was not a subtle statistical whisper; it was a loud, graded, monotonic effect that strengthened with each step up the reconstruction ladder.

Objective metrics alone, however, do not decide what radiologists actually see. A reconstruction can lower noise on paper while smearing away the fine reticular detail of the lung interstitium, trading one diagnostic liability for another. To address this, two radiologists independently reviewed the image sets and scored them on a five-point Likert scale across four dimensions: perceived noise, structural sharpness, artifact burden, and overall diagnostic acceptability. The composite subjective score rose from 3.21 with filtered back projection to 4.68 at the highest KARL 3D level, indicating that human observers, not just pixel statistics, registered the improvement. Agreement between and within the readers was assessed rigorously: intraclass correlation coefficients for the objective noise measurements ranged from 0.970 to 0.983, and quadratic-weighted kappa values for the subjective summed scores reached 0.86 to 0.88 between readers and 0.79 to 0.96 within the same reader across sessions—figures that fall in the good-to-excellent range and lend credibility to the perceptual findings.

One of the study’s most clinically consequential questions concerned body size. Radiology has long known that larger patients absorb more X-rays, produce noisier images, and often benefit differently from noise-reduction algorithms. If iterative reconstruction’s benefit shrank in larger patients, protocols might need to be tailored by body habitus, complicating workflow. The researchers therefore computed each patient’s effective diameter and tested whether it correlated with the degree of noise reduction. It did not. Pearson correlation coefficients hovered between 0.07 and 0.14, all statistically insignificant, and subgroup comparisons using the Mann-Whitney U test found no difference between body-size groups, with P values between 0.70 and 0.75. Within the range of patient sizes studied, the denoising effect of KARL 3D proved remarkably uniform.

That uniformity has practical implications. It suggests that a single reconstruction level can plausibly serve a broad patient population without per-patient adjustment, simplifying protocol design and reducing one more source of variability in a field already crowded with scanner models, kernel choices, and window settings. It also strengthens the argument for within-subject designs in imaging research more broadly: by holding anatomy and dose constant, the study isolated the algorithm’s contribution with a precision that cross-sectional comparisons rarely achieve.

The study is not without boundaries worth acknowledging. Thirty-six patients constitute a modest sample, all scanned on a single scanner model at a single institution with one dose level and one lung-window kernel. The retrospective design, approved by the institutional ethics committee with a waiver of informed consent, means the findings describe what the algorithm does under one specific set of conditions rather than guaranteeing identical behavior elsewhere. The authors themselves frame the conclusion carefully: KARL 3D produces a graded, level-dependent reduction in lung-window noise, and the choice of level should be guided by the clinical task. That last clause is the quiet heart of the paper. Higher levels progressively cleaned the images and improved subjective quality, but the optimal setting for detecting a subtle ground-glass opacity may differ from the optimal setting for evaluating a known nodule’s margins, and the study provides the gradient map rather than a single prescription.

The broader context makes the work timely. Iterative reconstruction has swept through clinical CT over the past decade as vendors raced to deliver lower radiation doses without sacrificing image quality, replacing filtered back projection in many departments. Yet the proliferation of proprietary algorithms—each with its own strength scale, its own noise texture, and its own trade-offs between smoothness and resolution—has left radiologists navigating a patchwork of incomparable settings. Studies that rigorously characterize one algorithm’s dose-response curve, using objective measurements and blinded observer scores with validated agreement statistics, are the raw material from which evidence-based protocol decisions are built. This paper adds a well-documented data point for a widely deployed Chinese scanner platform, where such characterization has been comparatively sparse.

For patients, the significance is indirect but real. Cleaner lung-window images at a fixed radiation dose mean radiologists can read chest CTs with less of the visual static that breeds uncertainty, and the demonstrated insensitivity to body size means that benefit extends across the patient spectrum rather than concentrating in the slim. For the imaging community, the message is that the KARL 3D dial behaves predictably: turn it up, and noise falls in a graded, statistically robust fashion, with observer agreement confirming that what the numbers promise is what the eyes receive. The next chapter—mapping which levels best serve which clinical questions, across more scanners, doses, and patient cohorts—remains open, but this study has drawn the baseline gradient with unusual clarity.

Subject of Research: Effect of KARL 3D iterative reconstruction levels on image noise in lung-window chest CT

Article Title: Gradient effect of KARL 3D iterative reconstruction on image noise in lung-window chest CT: a within-subject evaluation

Article References: Zhang, X., Peng, Y., Long, B., Ye, S., Zhu, Y., Han, R., & Peng, H. (2026). Gradient effect of KARL 3D iterative reconstruction on image noise in lung-window chest CT: a within-subject evaluation. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02922-y

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02922-y

Keywords: KARL 3D, iterative reconstruction, chest CT, lung window, image noise, signal-to-noise ratio, filtered back projection, effective diameter, radiology, medical imaging, computed tomography, image quality

The problem the team set out to solve is one that anyone working with cone-beam CT (CBCT) knows intimately. CBCT, originally developed for dental diagnostics, has become a workhorse of image-guided radiation therapy, interventional radiology and surgical navigation because it delivers high-resolution 3D images directly in the treatment room at lower radiation doses and faster scan times than conventional CT. But the traditional reconstruction method, the Feldkamp-Davis-Kress (FDK) algorithm, demands a large number of X-ray projections. When clinicians cut the number of projections to reduce dose — a critical safety goal — FDK produces noisy, artifact-riddled images.

Iterative reconstruction methods offer a way out. Instead of applying a single filtering step, they repeatedly refine an image estimate: the current guess is forward-projected to simulate what the scanner would measure, the simulated projections are compared with the actual measured data, and the difference is back-projected to update the image. Regularization terms, often based on the total variation (TV) norm, stabilize the solution by encouraging smooth tissue regions while preserving sharp anatomical edges. Algorithms such as ASD-POCS and its adaptive-weighted variant AwPCCS can produce strikingly good images from sparse, low-dose data — but only if their hyperparameters are set just right.

That is the catch. ASD-POCS alone carries eight tunable parameters, including the maximum number of iterations, the number of TV minimization steps, a data inconsistency tolerance, and several step-size and reduction factors spanning value ranges from a thousand candidate settings down to just ten. Even tiny deviations from the optimal combination can visibly degrade image quality. Manual tuning is tedious, time-consuming, and prone to observer variability, and any change in machine, organ, or radiation protocol typically forces clinicians to start over. Prior automated approaches have mostly optimized a single parameter, most often the regularization weight, and nearly all have relied on reference-based objective functions that assume a ground-truth reconstruction exists — an assumption that collapses in real clinical scenarios.

The new framework, called Search Space Aware Crow Search Algorithm (SSA-CSA), attacks both gaps at once. It is built on the crow search algorithm (CSA), a bio-inspired metaheuristic that models how crows hide food and follow flockmates to steal their caches. In the original CSA, each crow’s position encodes a candidate solution; here, each crow carries a vector of hyperparameter values, one per tunable parameter, allowing the entire parameter set of a reconstruction algorithm to be optimized simultaneously. The researchers reformulated all five core components of a metaheuristic — population initialization, exploitation, exploration, the balance between local and global search, and the fitness function — into a unified, reference-free pipeline.

Several technical innovations make the enhanced algorithm work. For initialization, the team developed a Chaotic Diagonal Linear Uniform (CDLU) scheme that combines diagonal stratified sampling with a deterministic sine chaos map, ensuring the initial flock of crows covers the enormous search space evenly without the ordering problems of standard diagonal sampling. For local search, crows now follow members of a dynamically shrinking ‘superior set’ of top performers identified via Pareto dominance, and random step factors are replaced with chaotic values. For global search, the crows gain ‘search space awareness’: a weight map is progressively built up around the best-performing regions of each parameter dimension, and exploration is steered by weighted random sampling rather than blind uniform randomness. The balance between exploitation and exploration is likewise made fitness-aware, so high-quality crows exploit while weaker ones explore, with the threshold tightening as the run progresses.

Equally important is the fitness function that guides the whole search. Because no reference image exists, the team designed a multi-objective, no-reference score combining the Signal-to-Noise Ratio (SNR), which captures noise reduction, with the high-frequency energy ratio (HFER), a Fourier-domain measure of image sharpness. The two metrics inherently conflict — smoother images are less noisy but blurrier — so the framework seeks the best trade-off point via a weighted sum. Ablation experiments on the publicly available SophiaBeads microCT dataset showed that this SNR-plus-HFER fitness outperformed alternatives based on SNR alone, Laplacian variance, or HFER alone when judged by two state-of-the-art learning-based image quality models, CHILL@UK and RPI_AXIS, and correlated strongly with those models’ scores.

The evaluation was unusually thorough. The framework was tested on six datasets from four different imaging systems, including a Nikon Custom Bay CT scanner, a medPhoton ImagingRing m device, and Philips Allura and Azurion C-arms, spanning glass-bead phantoms, line-pair CatPhan tests, a 3D-printed anthropomorphic thorax, a brain phantom with an inserted metal screw, a reconfigurable ChaosCube phantom, and simulated projections from five real patient CT volumes in the LIDC-IDRI lung database. Across ASD-POCS, AwPCSD and the prior-image-based PICCS algorithm, SSA-CSA consistently beat both manual parameter setting and the original CSA, with average improvements of roughly 3.9 percent in fitness, 4.3 percent on CHILL@UK and 3.9 percent on RPI_AXIS. A blind reader study by an expert radiologist with twenty years of CT experience ranked the SSA-CSA reconstructions first and awarded them the highest quality scores, including a perfect 5 out of 5 for the PICCS brain reconstruction from only 20 percent of the full projection set.

Perhaps the most clinically significant finding concerns generalizability. When the optimal parameter set found for one lung CT volume was applied to other patients and to different dose levels — from 100 percent of projections down to a severely undersampled 25 percent — image quality remained consistently good, indicating that a single optimization run can serve an entire class of similar imaging tasks. The same held for six different configurations of the ChaosCube phantom. In benchmark tests against eleven other optimization algorithms across nine standard functions, SSA-CSA showed fast convergence, low run-to-run variability attributable to its deterministic initialization, and statistically significant superiority in most pairwise Wilcoxon comparisons, even under the deliberately constrained budget of 25 individuals and 30 iterations chosen to reflect real clinical hardware limits.

The authors are candid about limitations: like all metaheuristics, SSA-CSA cannot guarantee a global optimum, its fitness weights require some experimental tuning, and its search-space awareness adds computational overhead — though this is negligible next to the cost of the reconstructions themselves and is incurred only once, offline. Future work may replace the hand-crafted fitness with a learned model or swap the metaheuristic for deep reinforcement learning. For now, the framework offers something clinics have lacked: a practical, operator-independent way to squeeze the best possible image quality out of low-dose iterative reconstruction, no reference image required.

Subject of Research: Automatic hyperparameter optimization for iterative cone-beam CT image reconstruction using an enhanced crow search metaheuristic algorithm

Article Title: Reference-free framework for automatic parameter optimization in iterative reconstruction using a novel search-space-aware crow search algorithm

Article References: MohammadiNasab, P., Biguri, A., Steininger, P., Keuschnigg, P., Lamminger, L., Lach, A., Islam, S. M. R. S., Breger, A., Karner, C., Schönlieb, C.-B., Birkfellner, W., & Hatamikia, S. (2026). Reference-free framework for automatic parameter optimization in iterative reconstruction using a novel search-space-aware crow search algorithm. Results in Engineering, 32, Article 113147. https://doi.org/10.1016/j.rineng.2026.113147

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113147

Keywords: cone-beam CT, iterative reconstruction, crow search algorithm, metaheuristics, hyperparameter optimization, low-dose imaging, image quality assessment, total variation regularization, ASD-POCS, PICCS, medical imaging, radiation dose reduction

The research team set out to answer a deceptively simple question: when a CT scanner is pushed to its lowest operating settings, how much computational rescue is needed to bring the images back up to diagnostic standards? To find out, they enrolled 107 patients who were scheduled for coronary CT angiography and divided them into two independent groups. Group A, comprising 52 patients, underwent what the researchers call a dual-low protocol: the scanner operated at a reduced tube voltage of 70 kilovolt peaks, and the contrast agent was a low-concentration formulation of iomeprol containing 300 milligrams of iodine per milliliter, injected at 1.0 milliliter per kilogram of body weight at a flow rate of 5.0 milliliters per second. Group B, with 55 patients, served as the reference standard, scanned at the conventional tube voltage of 120 kilovolt peaks with a higher-concentration 350 milligrams-of-iodine-per-milliliter contrast agent. Both groups were scanned with the same automatic tube current modulation, meaning the scanner adjusted its output to the patient’s body habitus in real time.

The crucial difference between the groups lay not just in the acquisition but in what happened to the raw data afterward. Images from Group A were reconstructed four separate ways using the Advanced Modeled Iterative Reconstruction algorithm, known as ADMIRE, at strength levels 1, 3, and 5, alongside conventional filtered back projection, the classical mathematical method that has underpinned CT reconstruction for decades. Filtered back projection is fast and computationally cheap, but it amplifies the statistical noise that inevitably creeps into images when fewer X-ray photons reach the detector. Iterative reconstruction algorithms take a fundamentally different approach: they repeatedly simulate what the detector should have recorded given a candidate image, compare that simulation against the actual measurement, and refine the image until the discrepancy is minimized. ADMIRE, a third-generation implementation of this idea, models the geometry of the scanner and the statistical behavior of the noise itself, allowing it to suppress graininess far more aggressively than earlier iterative methods without smearing away genuine anatomical detail.

The dose savings achieved by the dual-low protocol were striking. The volume CT dose index, a standard measure of radiation output, averaged 12.28 milligrays in the low-voltage group compared with 44.44 milligrays in the conventional group, a reduction of 72.4 percent that was highly statistically significant. The iodine uptake, a measure of how much contrast medium the patient actually received, fell by 18.6 percent as well, thanks to the combination of the lower-concentration agent and the weight-adjusted injection volume. These are not marginal refinements; they represent the difference between a scan that contributes meaningfully to a patient’s cumulative lifetime radiation exposure and one that approaches the dose of a routine chest X-ray series. For the iodine side of the equation, the benefit accrues to the kidneys and to patients at risk of contrast-induced nephropathy, a serious concern in the many cardiac patients who also have diabetes or chronic kidney disease.

Reducing dose and contrast, however, is only half the story, because an image that is cheap to produce but unreadable is clinically worthless. The researchers therefore subjected every reconstruction to a battery of objective measurements. They sampled CT values, the density readings within the coronary arteries, to confirm that the vessels were still opacified with contrast. They calculated the signal-to-noise ratio within the coronary lumen and the contrast-to-noise ratio between the coronary artery and the surrounding epicardial fat, both of which quantify how distinguishable the vessel is from its background. They also measured image noise directly as the standard deviation of pixel values in the aortic root. What they found was a textbook demonstration of what iterative reconstruction can do: across the four reconstruction strengths in the dual-low group, the CT values barely changed, confirming that the underlying contrast information was intact, while the noise in the aortic root plummeted from a standard deviation of 69.71 with filtered back projection to 41.19 at the highest ADMIRE strength of 5.

That collapse in noise translated directly into better ratios and better pictures. As the reconstruction strength climbed from 0 to 5, both the coronary signal-to-noise ratio and the coronary-to-fat contrast-to-noise ratio improved markedly, with p-values below 0.001 across the comparisons. Two independent radiologists, blinded to the reconstruction parameters, then graded every image on a five-point Likert scale assessing diagnostic confidence, and their subjective scores rose in lockstep with the objective metrics. In other words, the improvements were not statistical artifacts visible only in the numbers; human expert readers could see and appreciate the difference. The agreement between the objective physics and the subjective radiological judgment is what gives the study its practical force, because regulatory approval and clinical adoption ultimately depend on whether radiologists trust the images in front of them.

Perhaps the most remarkable finding of the study is how the dual-low images stacked up against the conventional-dose reference. Images reconstructed at ADMIRE strength 5 did not merely match the standard 120 kilovolt, high-iodine scans; they beat them, showing better objective signal-to-noise and contrast-to-noise ratios and higher subjective quality scores. Images at the intermediate strength of 3 were statistically comparable to the conventional scans. This means that a patient can receive less than a third of the radiation dose and nearly a fifth less iodine, and still walk away with a scan that is at least as informative as the traditional examination. The intermediate strength 3 setting meets traditional diagnostic standards, while strength 5 offers further optimization for readers who want maximum vessel conspicuity, giving departments a tunable dial rather than a single fixed compromise.

The physics behind why low tube voltage pairs so naturally with iterative reconstruction deserves attention. Lowering the tube voltage from 120 to 70 kilovolts shifts the X-ray spectrum closer to the k-edge of iodine, the energy at which iodine absorbs X-rays most strongly, so the same amount of contrast dye produces a brighter signal in the vessel lumen. That is why the coronary CT values did not fall despite the reduced iodine concentration. The penalty of the low-voltage approach is that the photons are more readily absorbed by the patient’s tissues, so fewer reach the detector and the quantum noise rises. Filtered back projection would leave that noise glaringly visible, but a noise-modeling iterative algorithm can distinguish the random statistical fluctuation from true structure and remove much of the former while preserving the latter. The dual-low strategy is thus a kind of physical and computational judo: it uses the enhanced iodine signal of low kilovolts to justify less contrast, and uses the algorithm to cancel the noise that the low dose would otherwise impose.

The clinical implications extend well beyond the scanner suite. Coronary artery disease remains the leading cause of death worldwide, and as CT angiography is increasingly used both to diagnose chest pain and to screen asymptomatic risk groups, the aggregate population radiation burden becomes a genuine public health consideration. Meanwhile, contrast-associated acute kidney injury is a feared complication in the very patients most likely to need cardiac imaging. A validated protocol that simultaneously addresses both hazards, using hardware and software that are already installed in modern third-generation dual-source CT systems, could be adopted without new capital investment. The study was conducted under the Declaration of Helsinki with ethics approval and written informed consent from all participants, and it was supported by the Beijing Huikang Ren’ai Public Welfare Foundation, though the funder played no role in the design, analysis, or publication of the work.

There are, of course, caveats that temper the enthusiasm. The study population of 107 patients, while adequate for the comparisons made, came from a single center with a single scanner platform, and the ADMIRE algorithm is proprietary to one manufacturer, so other institutions using different reconstruction engines will need to establish their own equivalence points. Very high reconstruction strengths have historically been associated with a waxy, over-smoothed texture that some readers find unnatural, and while the radiologists in this study scored strength 5 images highest, broader reader panels and larger cohorts will help confirm that impression. Still, the core message stands on solid ground: the combination of 70 kilovolt acquisition, low-concentration iodine contrast, and high-strength iterative reconstruction is feasible in routine clinical practice, and it delivers images that meet or exceed the conventional standard at a fraction of the biological cost. For the millions of patients who undergo coronary CT angiography each year, that is a trade-off worth celebrating.

Subject of Research: Iterative reconstruction optimization for low-dose, low-iodine coronary CT angiography

Article Title: Optimizing image quality in low-dose, low-iodine coronary CT angiography: a comparative study of iterative reconstruction strengths

Article References: Deng, K., Yang, H., Jiang, B., Guo, X., Mei, X., Hu, J., & Situ, W. (2026). Optimizing image quality in low-dose, low-iodine coronary CT angiography: a comparative study of iterative reconstruction strengths. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02831-0

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02831-0

Keywords: coronary CT angiography, iterative reconstruction, ADMIRE, radiation dose reduction, iodine contrast, image quality, signal-to-noise ratio, contrast-to-noise ratio, 70 kVp, dual-source CT, medical imaging, cardiology