RPS 1201 - Technical advances in abdominal imaging

Author Block: L. Hieronymi¹, J. Lueckel¹, S. Skornitzke², N. Große Hokamp¹, D. Maintz¹; ¹Cologne/DE, ²Hamburg/DE
Purpose: To evaluate a prototype deep learning–based spectral reconstruction algorithm (SAI, Philips) for spectral detector dual-energy CT (sdDECT) in abdominal imaging. The algorithm aims to reduce image noise while preserving texture in both conventional and virtual monoenergetic images (VMI), addressing known limitations of low-keV VMI.
Methods or Background: This retrospective study included 67 patients undergoing contrast-enhanced abdominal sdDECT. Conventional images were reconstructed with hybrid-iterative (HI-R) and spectral algorithm (SAI-R). VMI (40–200keV) were generated accordingly (HI-VMI, SAI-VMI). Quantitative analysis involved 22 ROIs in the liver, pancreas, spleen, kidneys, psoas muscle, and fat. Signal-to-noise and contrast-to-noise ratios (SNR/CNR) were calculated. Two blinded radiologists compared randomized, patient-matched images (HI-R/VMI vs. SAI-R/VMI, 40–70keV) across liver, pancreas, and kidney, rating image quality, noise, texture, lesion conspicuity, and diagnostic confidence using a two-alternative forced choice design. Statistical analysis was performed with ANOVA and Tukey’s post hoc test.
Results or Findings: Attenuation was higher in low-keV VMI (40–60keV) compared to conventional reconstructions (p≤0.05). Noise increased with decreasing keV, with SAI-VMI showing lower noise than HI-VMI at all levels (e.g., muscle at 40keV: 14.39±4.88HU vs. 16.15±4.90HU; p≤0.05). Importantly, SAI-VMI demonstrated lower noise at low-keV compared to HI-VMI at higher keV, highlighting superior noise performance at lower energy levels (e.g., liver: SAI at 40keV 12.17±3.71HU vs. HI at 60keV 13.10±3.34HU; p≤0.05). SNR and CNR were significantly higher with SAI-R and SAI-VMI, particularly at low-keV (e.g. liver CNR at 40 keV: SAI-VMI 8.18±2.67 vs. HI-VMI 7.15±2.41; p≤0.05). Reader preference strongly favored SAI-R/VMI (averaged: HI-R 0.83%±1.31% vs. SAI-R 31.67%±4.97%; p≤0.05). SAI-VMI at 40keV was preferred over all other reconstructions (97.92%±2.36%, p≤0.05).
Conclusion: AI-based spectral reconstruction improves abdominal sdDECT by reducing noise and enhancing SNR/CNR, while further increasing diagnostic benefits of VMI.
Limitations: The study is limited by its retrospective design.
Funding for this study: This work was funded by Philips Healthcare. The funding source had no involvement in study design, collection or interpretation of data.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study is IRB-approved.

Author Block: N. Kawai, Y. Noda, T. Kaga, K. Matsuoka, S. OMATA, Y. Takai, M. Asano, H. Kato, M. Matsuo; Gifu/JP
Purpose: The combination of high-frequency reconstruction kernels and deep-learning image reconstruction (DLIR) may enhance structure edges while maintaining image quality in abdominal CT. This study aimed to investigate the feasibility of combining high-frequency reconstruction kernels and DLIR at high-strength level (DLIR-H) for improving visualization of the pancreas and tumor boundaries on pancreatic protocol CT.
Methods or Background: This retrospective study included 30 patients (median age, 75 years; 16 women; median body mass index, 22 kg/m²) who underwent pancreatic protocol CT for assessing pancreatic tumors from January 2024 to July 2024. Four image sets were reconstructed using DLIR-H in combination with either standard, bone, bone-plus, or lung kernels. Edge sharpness between the pancreas and retroperitoneal fat tissue (pancreas-to-fat) and between the pancreas and pancreatic ductal adenocarcinoma (pancreas-to-PDAC) was quantitatively assessed using edge rise slope (ERS) measurements. Two radiologists qualitatively examined the sharpness of the pancreas, tumor boundary, and overall image quality.
Results or Findings: Pancreas-to-fat ERS was greater in lung kernel images than in standard and bone kernel images (P = 0.001). Pancreas-to-PDAC ERS was greater in lung kernel images than in other kernel images (P < 0.001). Sharpness of the pancreas and tumor boundaries was better in lung kernel images than in other kernel images (P < 0.001 for both). Overall image quality in lung kernel images was comparable to the standard and superior to the bone kernel (P < 0.001).
Conclusion: The combination of lung kernel and DLIR-H in pancreatic protocol CT improves both quantitative and qualitative sharpness of the pancreas and tumor boundaries while maintaining the overall image quality.
Limitations: Limitations include small sample size, lower body mass index than Western populations, no diagnostic performance assessment, and restriction to one vendor and DLIR-H.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Medical Research Ethics Review Committee, Gifu University Graduate School of Medicine

Author Block: A. Y. Yitik, R. Ambros, S. Ba-Ssalamah, S. Pötter-Lang, N. Bastati-Huber, A. Kristic, V. Vetchy, U. I. Attenberger, A. Ba-Ssalamah; Vienna/AT
Purpose: To evaluate determinants of technical failure of 3T liver MR elastography (MRE) using a 2D spin-echo EPI sequence in a large tertiary cohort.
Methods or Background: This retrospective study included 2,731 consecutive patients (mean age 55 ± 6 years; 52% male) who underwent 3T MRE between January 2020 and December 2023. Exams were performed with a 2D spin-echo EPI sequence. Two radiologists independently analyzed MRE and anatomic images. Liver iron concentration (LIC) and proton density fat fraction (PDFF) were measured on a 3D multi-echo GRE sequence. Technical failure was defined as the absence of pixels with >95% confidence index and/or lack of visible shear waves. Logistic regression was used to identify predictors of failure.
Results or Findings: The overall failure rate was 8% (205/2,731). On univariate analysis, higher body mass index (BMI), liver iron deposition, massive ascites, T1-mapping values, MELD, APRI, and ALBI scores were significantly associated with failure (all p < 0.05). Age, sex, PDFF, and FIB-4 were not. Multivariable analysis identified BMI, liver iron deposition, and massive ascites as independent predictors. R2* demonstrated fair accuracy for predicting failure (AUC = 0.783, 95% CI: 0.694–0.872, p < 0.001) with an optimal threshold of ≥171 s⁻¹ at 3T.
Conclusion: The technical failure rate of 3T MRE using 2D spin-echo EPI was 8%. Massive ascites, iron overload, and elevated BMI were independent predictors of unsuccessful exams. Identifying these risk factors may guide patient selection and optimize MRE performance in clinical practice.
Limitations: This single-center retrospective study using 3T 2D SE-EPI MRE is subject to design and selection bias.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics approval number: 1759/2023

Author Block: N. Kawai¹, Y. Noda¹, T. Kaga¹, Y. Ueda², M. Yoneyama², M. Asano¹, S. Ido¹, K. Kajita¹, M. Matsuo¹; ¹Gifu/JP, ²Tokyo/JP
Purpose: This study aimed to evaluate whether slice selection gradient reversal (SSGR) and multi-shot EPI (ms-EPI) could improve image quality compared with conventional single-shot EPI (ss-EPI) with spectral attenuated inversion recovery (SPAIR) in rectal DWI.
Methods or Background: This prospective study included 24 participants (median age, 70 years; 16 men) who underwent rectal MRI at 3T between March and August 2025. DWI was acquired using three sequences: ss-EPI with SPAIR, ss-EPI with SSGR, and ms-EPI with SSGR (two shots). Acquisition time was equalized by adjusting the number of signal averages (NSA; 8 for ss-EPI and 4 for ms-EPI). Two radiologists qualitatively assessed sharpness, distortion, artifact, lesion conspicuity, and overall image quality using a four-point Likert scale. Quantitative analysis included signal-to-noise ratio (SNR), signal intensity ratio (SIR), and apparent diffusion coefficient (ADC) value for rectal cancer.
Results or Findings: Rectal cancer was confirmed in 18 participants (75%). Sharpness, lesion conspicuity, and overall image quality were significantly better in the order of ms-EPI with SSGR, ss-EPI with SSGR, and ss-EPI with SPAIR images (P < 0.001 for all). Distortion and artifact were significantly reduced in ms-EPI with SSGR than in the other two images (P < 0.001 for both). SNR was higher in ss-EPI with SSGR than the other two images (P < 0.001). SIR was significantly higher in the order of ms-EPI with SSGR, ss-EPI with SSGR, and ss-EPI with SPAIR (P < 0.001). No significant differences were observed in ADC value (P = 0.84).
Conclusion: SSGR improved image quality compared with SPAIR. With equal acquisition times, ms-EPI showed lower SNR but was superior in reducing distortion and artifacts.
Limitations: Small sample size, use of a single-vendor 3T system, and adjustment of NSA to equalize scan time.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Medical Research Ethics Review Committee, Gifu University Graduate School of Medicine

Author Block: J. T. d. S. d. Castro, H. d. J. Ferreira, D. C. Novais da Silva, D. Yamada, S. San Juan Dertkigil, F. Reis; Campinas - Sao Paulo/BR
Purpose: Reducing iodine dose while maintaining diagnostic enhancement is crucial for patient safety, cost control, and environmental responsibility. The purpose was to quantify the per-patient and cohort-level contrast/iodine savings achieved after implementing a simple weight-based dosing table (1.0 mL/kg) aligned with low tube potential (80–100 kVp), compared with a conventional fixed-dose protocol.
Methods or Background: This was a single-center, real-world cohort study (pre/post implementation) analyzing 562 consecutive multiphasic abdominal CTs. Groups were: Conventional (pre-implementation, fixed dose) and three weight tiers in the post-implementation phase: G1 (≤90 kg @ 80 kVp), G2 (∼91–120 kg @ 100 kVp), and G3 (≥120 kg, exploratory). A counterfactual baseline contrast volume was estimated for each post-implementation patient based on the Conventional group's median mL/kg for that weight band. Per-patient saving was calculated as baseline minus observed volume. Total cohort saving (L) and enhancement results (from Poster 1) were the primary outcomes.
Results or Findings: The new protocol delivered meaningful contrast savings across all groups: G1 (≤90 kg, n=232): Achieved the highest per-patient savings (mean 24.9 mL; median 27.2 mL). G2 (∼91–120 kg, n=38): Showed modest but clear savings (mean 5.6 mL; median 6.4 mL). Total Savings: The cohort yielded a total saving of 7.40 L (G1) and 0.27 L (G2) within the dataset. Crucially, these substantial volume reductions co-occurred with higher proportions of exams meeting target portal enhancement (110–130 HU) and IRHR≥50 HU, indicating no diagnostic compromise.
Conclusion: A pragmatic 1.0 mL/kg weight-based dosing table integrated with 80–100 kVp delivers significant iodine/contrast savings at scale while successfully sustaining or improving diagnostic enhancement targets. This approach supports cost stewardship and environmental responsibility alongside demonstrable quality gains.
Limitations: Single-center, pre/post implementation cohort study, limiting the generalizability of specific savings figures to other institutions or protocols.
Funding for this study: Not applicable.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The work has been submitted to the university's Research Ethics Committee (CEP) and is currently under review for evaluation and final opinion.

Author Block: A. Ciprani, T. Polidori, G. Tremamunno, D. De Santis, F. Virgili, F. Pacelli, M. Zerunian, D. Caruso, M. Francone; Roma/IT
Purpose: To evaluate and compare the effect on quantitative image quality of two different contrast medium (CM) injection protocols using conventional and virtual monoenergetic imaging (VMI) at 50keV.
Methods or Background: In this retrospective study, Group A (n=30) underwent abdominal CT with 700mgI/kg of CM based on lean body weight (LBW) using conventional images, while Group B (n=30) received 500mgI/kg of CM using spectral derived images at 50keV. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were measured in the liver, spleen, kidneys, pancreas, aorta, and portal vein during arterial and portal phases. Comparisons were performed with the Wilcoxon signed-rank test and Bonferroni correction, and effect sizes (Cohen’s d) were calculated.
Results or Findings: Despite the lower iodine dose, Group B showed higher image quality. In the arterial phase, pancreatic SNR increased from 25.4 to 40.7 (+60%) and aortic SNR from 18.2 to 46.0 (+153%, p<0.001). Pancreatic CNR increased from 15.1 to 24.2 (+60%) and aortic CNR from 20.5 to 52.1 (+154%, p<0.001). In the portal venous phase, liver SNR increased from 28.6 to 48.0 (+68%) and portal vein SNR from 22.4 to 38.5 (+72%), with CNR showing the same trend (all p<0.001). Although spectral reconstructions had slightly higher noise (8.3 vs. 7.5 HU, p<0.05), the overall increase in SNR and CNR was substantial. Effect size analysis confirmed these differences, with Cohen’s d >0.8 for most structures and >1.2 for aortic and hepatic measurements.
Conclusion: VMI at 50keV showed higher SNR and CNR than conventional reconstructions, despite lower iodine dose. Improvements were consistent in both arterial and portal phases, indicating that low-keV VMI enables iodine reduction without compromising image quality, with potential benefits for safety and diagnostic confidence.
Limitations: Single-center study with small sample size and no clinical impact assessment.
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: N. É. Balogh, B. K. Budai, A. I. Vigh, E. Tóth, Á. Szücs, A. Szijártó, P. Maurovich Horvat, I. Dudás; Budapest/HU
Purpose: We aimed to evaluate whether 40 keV virtual monoenergetic image (VMI) reconstructions
from photon-counting detector CT (PCD-CT) offer improved diagnostic utility over standard
70 keV in preoperative assessment of pancreatic ductal adenocarcinoma (PDAC).
Methods or Background: Three-phase contrast-enhanced PCD-CT scans (arterial, portal, and venous) were analyzed in
50 patients with histologically confirmed PDAC. Lesion diameters, image quality, and
diagnostic confidence (1–5 scale) were assessed at both 40 and 70 keV. Morphologic and
staging features were compared; moreover, one radiologist rated whether 40 keV reconstructions
aided interpretation.
Results or Findings: Lesion diameters did not differ significantly between 40 and 70 keV, though isolated cases
showed discrepancies up to 18 mm. Image quality and diagnostic confidence were
significantly higher at 40 keV across all phases (p < 0.001). Morphologic and staging features
were highly consistent, with T-stage altered in only three patients. Subjective assessment
indicated that 40 keV reconstructions were helpful in 86% of cases. The portal venous phase
was perceived as the most beneficial, with 40 keV aiding interpretation in 37/50 (74%)
patients, followed by the venous phase in 36/50 (72%) and the arterial phase in 21/50 (42%).
In 16/50 (32%) patients, the reconstructions were helpful across all three phases, and in an
additional 19/50 (38%) patients, in two phases.
Conclusion: 40 keV VMI reconstructions significantly improve subjective image quality and diagnostic
confidence in PDAC imaging, without compromising objective lesion measurements or
staging interpretation. Occasional large differences in tumor size and consistently positive
subjective feedback support the clinical utility of 40 keV, particularly in the portal venous
phase, highlighting its value in preoperative staging and surgical planning.
Limitations: This was a single-center study with a retrospective study design.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by the institutional ethics committee.

Author Block: F. Minonne, A. P. Micheletti, F. Giorgino, G. De Santis, S. Tibaldi, M. Moretti, R. Aroasio, M. Gatti, R. Faletti; Turin/IT
Purpose: Accurate preoperative estimation of donor liver weight is a critical step in optimizing organ allocation and surgical planning for liver transplantation. This study evaluated the feasibility of automated computed tomography (CT)–based volumetric analysis to predict liver weight and its correlation with clinical parameters.
Methods or Background: We conducted a retrospective, single-center observational study at the Città della Salute e della Scienza di Torino, including 103 deceased donors between 2014 and 2025 (age range: 13–84 years). Each subject underwent abdominal CT close to organ procurement for orthotopic liver transplantation (OLT). Donors were stratified into two groups according to imaging protocol: non-contrast CT and contrast-enhanced CT. Each group was further split into a training cohort, used to refine weight estimation algorithms, and a validation cohort, designed to test predictive accuracy.
Results or Findings: Liver segmentation was performed using the open-source software TotalSegmentator. Mean processing time was under two minutes per scan. Across both imaging groups, predicted liver weight closely approximated actual graft weight at explant, with a mean absolute error of ±0.14 kg, corresponding to about 9% of the true value. This degree of precision was consistent between non-contrast and contrast-enhanced acquisitions.
Conclusion: These findings support the feasibility of fully automated, convolutional neural network–based approaches for liver volumetry in transplantation. The proposed workflow demonstrated speed, reproducibility, and satisfactory accuracy. By reducing the time and variability associated with manual volumetric assessments, this method could improve efficiency in clinical practice and enhance the reliability of preoperative planning in liver transplantation.
Limitations: No
Funding for this study: No
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No Additional Information

Author Block: E. Petrash, K. Poliakov, M. Basova, M. Dugova, R. Gareeva, A. Zakharov, V. Gombolevskiy; Moscow/RU
Purpose: Splenic volumetry is important for splenomegaly detection. The aim is to assess agreement and robustness of spleen volume estimation methods on CT. Manual linear measurements used for volume estimation may be slice- and reader-dependent, whereas AI segmentation is expected to provide reproducible, protocol-independent volumes once the software version is fixed.
Methods or Background: Experienced radiologists independently measured craniocaudal length (L), maximal diameter (D), and thickness (T) on abdominal CT and computed volumes via the formula V =30+0.58× L×D×T. The same scans were processed by a CNN-based tool Spleen-IRA which automatically extracted dimensions and volumes. Algorithm-generated masks were validated by two radiologists with more than 12 years experience, disagreements were reviewed to ensure quality. Verified AI measurements were designated as reference for splenic volume.
Results or Findings: Statistical tests demonstrated significant differences between readers and between reader-derived and AI-derived spleen volumes (p < 10⁻⁷), underscoring the variability of manual measurements compared to the reproducibility of AI-based volumetry.
Conclusion: Manual spleen volumetry is time-consuming and inter-reader variable. Moreover, the observed expert disagreement suggests that the linear-measurement methodology itself is imperfect (landmarking and slice-selection dependent). AI segmentation delivers reader-independent spleen volumes without geometric approximations. With fixed inference settings, AI outputs are consistent on identical inputs. We therefore recommend AI segmentation as the reference method, and using the curated, quality-controlled mask dataset to audit legacy formulae and develop protocol-specific alternatives for settings where AI cannot be deployed.
Limitations: None
Funding for this study: IRA labs
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: