Low-energy dual-energy CT combined with deep learning reconstruction may make hypovascular liver lesions easier to see while allowing less iodinated contrast to be used. A phantom study from Lausanne University Hospital, published in European Radiology Experimental, tested virtual monoenergetic images at several energy levels and across three simulated body sizes. Images reconstructed at 40 keV with high-strength deep learning image reconstruction gave the best lesion detectability in every tested condition. The model also suggested that contrast volume could be reduced by at least 31.3%, although the possible reduction became smaller as simulated body size increased and still requires clinical confirmation.
Must Read: CT Contrast Enhancement Depends on Multiple Factors
A Phantom Model Based on Clinical Imaging
The experiment used an anthropomorphic abdominal phantom set up to represent small, medium and large body sizes. The three configurations represented underweight, normal-weight and overweight categories. Its design and scanning conditions were based on data from 90 thoraco-abdomino-pelvic dual-energy CT examinations performed for oncological follow-up in patients with confirmed liver metastases. Extension rings were added to reproduce differences in soft tissue and adipose tissue.
Special modules measured image contrast, noise and spatial resolution. Iodine concentrations inside the phantom were chosen to match attenuation values seen in liver tissue and hypovascular metastases during portal venous imaging. Simulated lesions measured between 5 and 10 mm, reflecting the sizes found in the clinical examinations used to build the model. Scan dose and field of view were adjusted for each phantom size. The phantom was scanned with the same clinical dual-energy CT protocol.
Virtual monoenergetic images were reconstructed from 40 to 70 keV using high-strength deep learning image reconstruction and adaptive statistical iterative reconstruction at 50% strength. A mathematical observer estimated lesion detectability from contrast, noise, spatial resolution and lesion size. The possible reduction in iodinated contrast was then calculated by comparing the best energy level with the reference level of 70 keV. This produced a theoretical estimate only, because the study did not directly test lower contrast doses.
Forty-keV Images Improve Lesion Visibility
Image contrast increased as the energy level fell and was highest at 40 keV. At this level, contrast was about four times higher than at 70 keV. It remained similar between the two reconstruction methods and across all three phantom sizes. Body size therefore had little effect on contrast enhancement, although it had a clearer effect on spatial resolution and lesion visibility.
Noise was also highest at 40 keV, but deep learning reconstruction reduced it substantially compared with iterative reconstruction. Noise fell by 44% at 40 keV and by 33% at 70 keV. At 40 keV, noise with deep learning reconstruction was similar to that seen at 60 keV with iterative reconstruction. Iterative reconstruction gave slightly better spatial resolution, but this did not stop the deep learning method from producing better overall lesion detectability.
All simulated lesions were technically detectable in every tested condition. Detectability was always highest at 40 keV, improved as lesion size increased and declined as phantom size increased. Larger lesions remained easier to identify than smaller lesions under every condition. At 40 keV, deep learning reconstruction improved detectability over iterative reconstruction by 62% in the small phantom, 71% in the medium phantom and 33% in the large phantom. The advantage was therefore strongest in the small and medium configurations.
Body Size Limits Contrast Reduction
The estimated scope for reducing iodinated contrast depended on both body size and lesion size. For 5-mm lesions, the theoretical reduction was 36.5% in the small phantom, 41.9% in the medium phantom and 31.3% in the large phantom. Larger lesions allowed greater estimated reductions, while deep learning reconstruction consistently offered more scope for reduction than iterative reconstruction.
The lower margin in the large phantom reflected the fall in lesion detectability as simulated body size increased. Although contrast enhancement stayed broadly stable, larger phantom dimensions reduced spatial resolution and lesion visibility. The findings therefore support contrast optimisation that takes body habitus into account rather than applying the same reduction to every patient. Lower contrast use in repeated oncological imaging may also reduce contrast-related complications, healthcare costs and the release of iodinated contrast into aquatic ecosystems.
The estimates remain theoretical and require clinical confirmation. Contrast volume was not reduced during the experiment, and the calculations came from a detectability index under controlled phantom conditions. Human tissues are more varied, lesions differ in location and appearance, and very small lesions may be harder to detect in practice. The work also used one scanner platform and one radiation dose for each body-size group. Clinical studies across different scanners and iodine doses are needed before routine use.
Combining 40-keV virtual monoenergetic imaging with high-strength deep learning reconstruction improved the visibility of simulated hypovascular liver lesions across all tested body sizes. The phantom model supported a theoretical iodinated contrast reduction of at least 31.3%, with less scope in the largest configuration and more scope for larger lesions. However, lower contrast doses were not directly tested. Clinical studies must confirm whether these reductions can preserve lesion detection across different patients, scanners and imaging conditions before the approach is introduced into routine practice.
Source: European Radiology Experimental
References:
Gulizia M, Dromain C, Haefliger L et al. (2026) Optimization of hypovascular liver lesion detectability in dual-energy CT using deep learning image reconstruction: a phantom study for potential iodine dose reduction. Eur Radiol Exp; 10, 104.
