Yankai Meng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yankai Meng has a total of 141 co-authors. Here are the top co-authors by number of shared publications:
Kai Xu
Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, China
21 shared publicationsJuan Long
Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, China
18 shared publicationsAiyun Sun
CT Imaging Research Center, GE Healthcare China, Shanghai, China
17 shared publicationsXiaohan Liu
Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, P.R. China
17 shared publicationsChunfeng Hu
Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, China
13 shared publicationsHe Zhang
Department of Radiology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, China
13 shared publications
Latest Publications
- Open Access
- 25 August 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Development and validation of a machine learning nomogram model for differentiating mycoplasma pneumoniae coinfection with other pathogens in children
- 25 August 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Open AccessDevelopment and validation of a machine learning nomogram model for differentiating mycoplasma pneumoniae coinfection with other pathogens in children
- 10 August 2026
- Published by Wolters Kluwer Health in Journal of Thoracic Imaging
- Vol. 41 (6)
High-pitch Low-dose Chest Computed Tomography With Calcium-aware Reconstruction for Coronary Artery Calcium Scoring
Purpose: To evaluate the accuracy, radiation dose, and clinical feasibility of a high-pitch low-dose chest CT (Sa36LDCT) scanning protocol with calcium-aware reconstruction for coronary artery calcium scoring (CAC). Materials and Methods: In this prospective study, 90 patients underwent both standard CAC scanning (Qr36CACS, 120 kVp) and Sa36LDCT (70 to 120 kVp, high-pitch, ATVS). CAC scoring was quantified as Agatston score, calcium volume, and equivalent mass. Risk stratification was assessed using Agatston categories. Agreement between protocols was evaluated using intraclass correlation coefficients, Bland–Altman analysis, and weighted kappa. Subgroup analyses were performed according to heart rate (≤75 vs. >75 bpm). Radiation dose and subjective image quality were also compared. Results: CAC metrics showed no significant difference between protocols, with excellent agreement (ICC = 0.983 to 0.996). Subgroup analysis revealed that heart rate did not significantly influence CAC quantification or risk classification, with Agatston score ICC values of 0.98 for ≤75 bpm and 0.99 for >75 bpm, demonstrating similar accuracy in both subgroups. Radiation dose reduction was consistently observed across both subgroups, with Sa36LDCT reducing effective radiation dose by 62% (0.74 vs. 0.28 mSv, P < 0.001), without compromising image quality. Risk stratification based on Agatston categories showed near-perfect consistency across heart rate subgroups (weighted k = 0.957). Conclusion: High-pitch Sa36LDCT provides accurate and reliable CAC scoring equivalent to standard CAC scoring, with substantial radiation dose reduction and preserved image quality. The protocol’s robustness across heart rate subgroups demonstrates its potential as a safe and practical approach for combined pulmonary and cardiovascular screening, pending validation in larger, more diverse populations.
- 10 August 2026
- Published by Wolters Kluwer Health in Journal of Thoracic Imaging
- Vol. 41 (6)
High-pitch Low-dose Chest Computed Tomography With Calcium-aware Reconstruction for Coronary Artery Calcium Scoring
Purpose: To evaluate the accuracy, radiation dose, and clinical feasibility of a high-pitch low-dose chest CT (Sa36LDCT) scanning protocol with calcium-aware reconstruction for coronary artery calcium scoring (CAC). Materials and Methods: In this prospective study, 90 patients underwent both standard CAC scanning (Qr36CACS, 120 kVp) and Sa36LDCT (70 to 120 kVp, high-pitch, ATVS). CAC scoring was quantified as Agatston score, calcium volume, and equivalent mass. Risk stratification was assessed using Agatston categories. Agreement between protocols was evaluated using intraclass correlation coefficients, Bland–Altman analysis, and weighted kappa. Subgroup analyses were performed according to heart rate (≤75 vs. >75 bpm). Radiation dose and subjective image quality were also compared. Results: CAC metrics showed no significant difference between protocols, with excellent agreement (ICC = 0.983 to 0.996). Subgroup analysis revealed that heart rate did not significantly influence CAC quantification or risk classification, with Agatston score ICC values of 0.98 for ≤75 bpm and 0.99 for >75 bpm, demonstrating similar accuracy in both subgroups. Radiation dose reduction was consistently observed across both subgroups, with Sa36LDCT reducing effective radiation dose by 62% (0.74 vs. 0.28 mSv, P < 0.001), without compromising image quality. Risk stratification based on Agatston categories showed near-perfect consistency across heart rate subgroups (weighted k = 0.957). Conclusion: High-pitch Sa36LDCT provides accurate and reliable CAC scoring equivalent to standard CAC scoring, with substantial radiation dose reduction and preserved image quality. The protocol’s robustness across heart rate subgroups demonstrates its potential as a safe and practical approach for combined pulmonary and cardiovascular screening, pending validation in larger, more diverse populations.
- 13 July 2026
- Published by Springer Nature in Abdominal Radiology
Integrated dual-energy CT spectral kinetics and subregional habitat phenotyping for preoperative prediction of perineural invasion in gastric adenocarcinoma.
Perineural invasion (PNI) is a key prognostic determinant in gastric adenocarcinoma but remains radiologically occult on conventional CT. To develop and validate an integrated model combining dual-energy CT (DECT)-derived functional spectral slopes and subregional habitat signatures for the preoperative prediction of PNI. In this retrospective study, 175 patients (median age, 66 years; interquartile range, 58-73 years; 131 men) with gastric adenocarcinoma who underwent preoperative DECT between 2023 and 2025 were evaluated. Patients were randomized into a training cohort (n = 122) and an internal test cohort (n = 53). Tumor subregions were parcellated into three biologically distinct habitats using voxel-wise clustering. Stable habitat features (ICC ≥ 0.75), three-phase spectral slopes, and a Habitat-Graph Neural Network (H-GNN)-derived Topological Interaction Score (TIS)-capturing spatial adjacency relationships between habitat subregions-were collectively submitted as candidate predictors to LASSO regression for construction of the final Spectral-Habitat Signature. Diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC). Clinical utility was assessed via decision curve analysis. PNI was pathologically confirmed in 46.9% (82 of 175) of patients. The integrated Spectral-Habitat model achieved an AUC of 0.832 (95% CI: 0.760, 0.898) in the training cohort and 0.758 (95% CI: 0.666, 0.845) in the internal test cohort. A stratified 5-fold cross-validation demonstrated the stability of the framework, yielding a mean AUC of 0.765 (SD, 0.084). The combined model significantly outperformed the extracellular volume (ECV) model (AUC, 0.758 vs. 0.544; P < .001). Equilibrium-phase spectral slope emerged as a decisive functional biomarker. Model calibration was excellent in the test cohort (P = .560, Hosmer-Lemeshow test). Decision curve analysis demonstrated a substantial net clinical benefit across threshold probabilities of 10%-80%. In this retrospective single-center study, an integrated model leveraging functional spectral kinetics and subregional spatial heterogeneity showed potential for the preoperative stratification of occult perineural invasion. However, these preliminary findings derived from a limited sample size require further validation in larger, multi-institutional prospective cohorts before clinical implementation in surgical planning.
- 13 July 2026
- Published by Springer Nature in Abdominal Radiology
Integrated dual-energy CT spectral kinetics and subregional habitat phenotyping for preoperative prediction of perineural invasion in gastric adenocarcinoma.
Perineural invasion (PNI) is a key prognostic determinant in gastric adenocarcinoma but remains radiologically occult on conventional CT. To develop and validate an integrated model combining dual-energy CT (DECT)-derived functional spectral slopes and subregional habitat signatures for the preoperative prediction of PNI. In this retrospective study, 175 patients (median age, 66 years; interquartile range, 58-73 years; 131 men) with gastric adenocarcinoma who underwent preoperative DECT between 2023 and 2025 were evaluated. Patients were randomized into a training cohort (n = 122) and an internal test cohort (n = 53). Tumor subregions were parcellated into three biologically distinct habitats using voxel-wise clustering. Stable habitat features (ICC ≥ 0.75), three-phase spectral slopes, and a Habitat-Graph Neural Network (H-GNN)-derived Topological Interaction Score (TIS)-capturing spatial adjacency relationships between habitat subregions-were collectively submitted as candidate predictors to LASSO regression for construction of the final Spectral-Habitat Signature. Diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC). Clinical utility was assessed via decision curve analysis. PNI was pathologically confirmed in 46.9% (82 of 175) of patients. The integrated Spectral-Habitat model achieved an AUC of 0.832 (95% CI: 0.760, 0.898) in the training cohort and 0.758 (95% CI: 0.666, 0.845) in the internal test cohort. A stratified 5-fold cross-validation demonstrated the stability of the framework, yielding a mean AUC of 0.765 (SD, 0.084). The combined model significantly outperformed the extracellular volume (ECV) model (AUC, 0.758 vs. 0.544; P < .001). Equilibrium-phase spectral slope emerged as a decisive functional biomarker. Model calibration was excellent in the test cohort (P = .560, Hosmer-Lemeshow test). Decision curve analysis demonstrated a substantial net clinical benefit across threshold probabilities of 10%-80%. In this retrospective single-center study, an integrated model leveraging functional spectral kinetics and subregional spatial heterogeneity showed potential for the preoperative stratification of occult perineural invasion. However, these preliminary findings derived from a limited sample size require further validation in larger, multi-institutional prospective cohorts before clinical implementation in surgical planning.
- 11 June 2026
- Published by Springer Nature in Japanese Journal of Radiology
- Vol. 44 (10) , 1-11
Development and performance of a vascular inflammation index for spontaneous isolated superior mesenteric artery dissection based on automated radial spatial PVAT analysis
Purpose Spontaneous isolated superior mesenteric artery dissection (SISMAD) is traditionally diagnosed via CT angiography (CTA) structural features, but localized biological inflammation remains poorly characterized. We aimed to characterize the radial spatial gradient of perivascular adipose tissue (PVAT) in SISMAD and evaluate a composite Vascular Inflammation Index (VII) as an auxiliary indicator of local inflammatory activity. Materials and methods In this retrospective study (2016–2025), 55 patients with SISMAD and 110 symptomatic controls were evaluated. PVAT volume, mean attenuation, and standard deviation (SD) were quantified across cumulative radial shells (0–1 to 0–5 mm) from the SMA wall. A composite VII, integrating volume burden, expansion kinetics, and tissue texture, was developed. Multivariable logistic regression and bootstrap analysis (1000 iterations) assessed independent predictors and incremental diagnostic value. Results Participants (mean age, 50.4 years ± 15.9) included SISMAD patients who had higher white blood cell counts than controls (10.7 ± 5.2 vs. 8.7 ± 4.0 × 109/L; P = 0.017). PVAT volume was significantly higher in SISMAD (0–5 mm: 10.1 mm3 ± 5.1 vs. 6.8 mm3 ± 4.0; P < 0.001). SISMAD patients exhibited aggressive radial expansion (expansion ratio: 12.5 vs. 8.6; P = 0.032) and lower 3-mm heterogeneity (22.8 vs. 24.1 HU; P = 0.046), suggesting an edema-induced homogenization effect. The VII achieved an AUC of 0.800 (95% CI 0.723, 0.878) for the identification of SISMAD-associated periarterial involvement and was the independent predictor of SISMAD (OR, 3.23; P < 0.001). Adding VII to the clinical model (Age, Gender, BMI, and WBC) significantly improved performance (AUC, 0.892 vs. 0.857; P = 0.038; increment, 0.035). A seven-day inflammatory plateau was identified post-onset. Conclusion Automated radial spatial profiling of PVAT reveals a distinct volumetric and kinetic inflammatory signature in SISMAD. The integrated VII provides a quantitative biological assessment that complements structural CTA diagnosis.
- 11 June 2026
- Published by Springer Nature in Japanese Journal of Radiology
- Vol. 44 (10) , 1-11
Development and performance of a vascular inflammation index for spontaneous isolated superior mesenteric artery dissection based on automated radial spatial PVAT analysis
Purpose Spontaneous isolated superior mesenteric artery dissection (SISMAD) is traditionally diagnosed via CT angiography (CTA) structural features, but localized biological inflammation remains poorly characterized. We aimed to characterize the radial spatial gradient of perivascular adipose tissue (PVAT) in SISMAD and evaluate a composite Vascular Inflammation Index (VII) as an auxiliary indicator of local inflammatory activity. Materials and methods In this retrospective study (2016–2025), 55 patients with SISMAD and 110 symptomatic controls were evaluated. PVAT volume, mean attenuation, and standard deviation (SD) were quantified across cumulative radial shells (0–1 to 0–5 mm) from the SMA wall. A composite VII, integrating volume burden, expansion kinetics, and tissue texture, was developed. Multivariable logistic regression and bootstrap analysis (1000 iterations) assessed independent predictors and incremental diagnostic value. Results Participants (mean age, 50.4 years ± 15.9) included SISMAD patients who had higher white blood cell counts than controls (10.7 ± 5.2 vs. 8.7 ± 4.0 × 109/L; P = 0.017). PVAT volume was significantly higher in SISMAD (0–5 mm: 10.1 mm3 ± 5.1 vs. 6.8 mm3 ± 4.0; P < 0.001). SISMAD patients exhibited aggressive radial expansion (expansion ratio: 12.5 vs. 8.6; P = 0.032) and lower 3-mm heterogeneity (22.8 vs. 24.1 HU; P = 0.046), suggesting an edema-induced homogenization effect. The VII achieved an AUC of 0.800 (95% CI 0.723, 0.878) for the identification of SISMAD-associated periarterial involvement and was the independent predictor of SISMAD (OR, 3.23; P < 0.001). Adding VII to the clinical model (Age, Gender, BMI, and WBC) significantly improved performance (AUC, 0.892 vs. 0.857; P = 0.038; increment, 0.035). A seven-day inflammatory plateau was identified post-onset. Conclusion Automated radial spatial profiling of PVAT reveals a distinct volumetric and kinetic inflammatory signature in SISMAD. The integrated VII provides a quantitative biological assessment that complements structural CTA diagnosis.
- 22 April 2026
- Published by Springer Nature in Clinical Neuroradiology
- Vol. 36 (3) , 1-15
Plaque-Level Machine Learning Prediction of Intraplaque Hemorrhage in Carotid Arteries Using Computed Tomography Angiography
Background Carotid artery plaques, especially those with intraplaque hemorrhage (IPH), are significant contributors to ischemic stroke. Although high-resolution magnetic resonance vessel-wall imaging (HR-MR-VWI) is the gold standard for assessing plaque vulnerability, its limited availability and high cost pose challenges. Computed tomography angiography (CTA) offers a more accessible, non-invasive alternative, but its ability to detect IPH and assess plaque instability remains underexplored. Machine learning techniques have shown promise in improving the prediction of carotid plaque vulnerability using CTA. Objectives This study aimed to develop and validate a machine learning model using CTA to predict IPH in carotid plaques. The model integrated key imaging features, including plaque composition, vascular lumen geometry, and perivascular adipose tissue (PVAT), with HR-MR-VWI serving as the reference standard. The goal was to evaluate the model’s potential for non-invasive plaque vulnerability prediction, particularly in clinical settings where MRI is not available. Methods A retrospective analysis was conducted on patients who underwent both carotid CTA and HR-MR-VWI within one month. Key plaque features, including composition, vascular lumen geometry, and PVAT, were extracted from CTA. The dataset was split into training (70%) and validation (30%) sets. Feature selection was performed using LASSO regression, followed by model development with logistic regression, random forest, XGBoost, and support vector machine (SVM). Hyperparameter tuning was performed using 10-fold cross-validation. Model performance was assessed using AUC, ROC curves, calibration curves, precision-recall curves, and confusion matrices. SHAP analysis was employed to evaluate the importance of each feature. Results In the validation cohort, the Random Forest model achieved an AUC of 0.679, sensitivity of 86.0%, specificity of 45.6%, accuracy of 63.0%, positive predictive value (PPV) of 54.4%, and negative predictive value (NPV) of 81.2%. Feature selection using LASSO regression identified perivascular adipose tissue (PVAT) attenuation, maximum diameter stenosis (MDS), and fibrotic volume (FV) as the most important predictors. SHAP analysis confirmed MDS as the most influential feature, followed by FV and PVAT. The model demonstrated good calibration, with predicted probabilities aligning closely with observed outcomes. Decision Curve Analysis (DCA) showed that the Random Forest model provided the highest net benefit at higher decision thresholds, supporting its clinical potential for predicting plaque vulnerability. Conclusion This study demonstrates the potential of a CTA-based machine learning model for predicting carotid plaque vulnerability, with PVAT, MDS, and FV as key predictors. While the model shows good sensitivity, its moderate specificity suggests the need for further refinement, particularly by incorporating additional clinical and morphological data. Given its high sensitivity and negative predictive value, this model is positioned as a non-invasive screening and triage tool to identify patients who may benefit from subsequent HR-MR-VWI, rather than a standalone definitive diagnostic classifier. Future research should focus on multi-center validation and the integration of clinical data to improve accuracy and clinical applicability.
- 22 April 2026
- Published by Springer Nature in Clinical Neuroradiology
- Vol. 36 (3) , 1-15
Plaque-Level Machine Learning Prediction of Intraplaque Hemorrhage in Carotid Arteries Using Computed Tomography Angiography
Background Carotid artery plaques, especially those with intraplaque hemorrhage (IPH), are significant contributors to ischemic stroke. Although high-resolution magnetic resonance vessel-wall imaging (HR-MR-VWI) is the gold standard for assessing plaque vulnerability, its limited availability and high cost pose challenges. Computed tomography angiography (CTA) offers a more accessible, non-invasive alternative, but its ability to detect IPH and assess plaque instability remains underexplored. Machine learning techniques have shown promise in improving the prediction of carotid plaque vulnerability using CTA. Objectives This study aimed to develop and validate a machine learning model using CTA to predict IPH in carotid plaques. The model integrated key imaging features, including plaque composition, vascular lumen geometry, and perivascular adipose tissue (PVAT), with HR-MR-VWI serving as the reference standard. The goal was to evaluate the model’s potential for non-invasive plaque vulnerability prediction, particularly in clinical settings where MRI is not available. Methods A retrospective analysis was conducted on patients who underwent both carotid CTA and HR-MR-VWI within one month. Key plaque features, including composition, vascular lumen geometry, and PVAT, were extracted from CTA. The dataset was split into training (70%) and validation (30%) sets. Feature selection was performed using LASSO regression, followed by model development with logistic regression, random forest, XGBoost, and support vector machine (SVM). Hyperparameter tuning was performed using 10-fold cross-validation. Model performance was assessed using AUC, ROC curves, calibration curves, precision-recall curves, and confusion matrices. SHAP analysis was employed to evaluate the importance of each feature. Results In the validation cohort, the Random Forest model achieved an AUC of 0.679, sensitivity of 86.0%, specificity of 45.6%, accuracy of 63.0%, positive predictive value (PPV) of 54.4%, and negative predictive value (NPV) of 81.2%. Feature selection using LASSO regression identified perivascular adipose tissue (PVAT) attenuation, maximum diameter stenosis (MDS), and fibrotic volume (FV) as the most important predictors. SHAP analysis confirmed MDS as the most influential feature, followed by FV and PVAT. The model demonstrated good calibration, with predicted probabilities aligning closely with observed outcomes. Decision Curve Analysis (DCA) showed that the Random Forest model provided the highest net benefit at higher decision thresholds, supporting its clinical potential for predicting plaque vulnerability. Conclusion This study demonstrates the potential of a CTA-based machine learning model for predicting carotid plaque vulnerability, with PVAT, MDS, and FV as key predictors. While the model shows good sensitivity, its moderate specificity suggests the need for further refinement, particularly by incorporating additional clinical and morphological data. Given its high sensitivity and negative predictive value, this model is positioned as a non-invasive screening and triage tool to identify patients who may benefit from subsequent HR-MR-VWI, rather than a standalone definitive diagnostic classifier. Future research should focus on multi-center validation and the integration of clinical data to improve accuracy and clinical applicability.
- Open Access
- 22 April 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Enhancing mesenteric vascular imaging with dual-energy CTA: a comparison of DLIR and ASIR-V using low contrast agent and low radiation dose protocols
Timely diagnosis of mesenteric vascular diseases, especially acute mesenteric ischemia (AMI) due to embolism in the superior mesenteric artery (SMA), is cr
- 22 April 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Open AccessEnhancing mesenteric vascular imaging with dual-energy CTA: a comparison of DLIR and ASIR-V using low contrast agent and low radiation dose protocols
Timely diagnosis of mesenteric vascular diseases, especially acute mesenteric ischemia (AMI) due to embolism in the superior mesenteric artery (SMA), is cr
- 10 April 2026
- Published by Springer Nature in Neuroradiology
- Vol. 68 (7) , 1963-1978
Plaque-level machine-learning prediction of carotid plaque vulnerability on computed tomography angiography
- 10 April 2026
- Published by Springer Nature in Neuroradiology
- Vol. 68 (7) , 1963-1978
Plaque-level machine-learning prediction of carotid plaque vulnerability on computed tomography angiography
- 9 April 2026
- Published by Elsevier in Magnetic Resonance Imaging
- Vol. 131, 110681
Ultrashort echo time T2⁎ mapping of early hip osteoarthritis: Acetabular predominance and loss of joint homeostasis
- 9 April 2026
- Published by Elsevier in Magnetic Resonance Imaging
- Vol. 131, 110681
Ultrashort echo time T2⁎ mapping of early hip osteoarthritis: Acetabular predominance and loss of joint homeostasis
- 6 April 2026
- Published by Springer Nature in Abdominal Radiology
- Vol. 51 (10) , 4820-4833
Deep learning reconstruction for 40-keV virtual monoenergetic CT of colon cancer: evaluation of image quality and edge sharpness
Virtual monoenergetic imaging (VMI) at 40 keV improves iodine attenuation in colon cancer CT but is constrained by severe image noise. Deep learning image reconstruction (DLIR) may address this limitation, but its effect on anatomical edge preservation across multiple targets requires investigation. To evaluate the impact of DLIR on objective and subjective image quality of 40-keV VMIs in colon adenocarcinoma, with emphasis on the trade-off between noise reduction and edge definition. In this completed retrospective study (patient enrollment window: May 2024 to February 2025), 60 patients (mean age, 62.8 years ± 15.1; 34 men) with confirmed colon adenocarcinoma underwent dual-energy CT using a low-iodine protocol (1.0 mL/kg). Portal venous phase data were reconstructed at 40 keV using adaptive statistical iterative reconstruction-V (ASIR-V) 50%, medium-strength DLIR (DLIR-M), and high-strength DLIR (DLIR-H). Objective metrics, including contrast-to-noise ratio (CNR), edge rise distance (ERD), and edge rise slope (ERS), were measured. Two radiologists independently scored five qualitative categories, including regional lymph node visualization, using a 5-point Likert scale. DLIR-H yielded the lowest image noise and highest CNR across all anatomical targets compared with DLIR-M and ASIR-V 50% (all P < 0.001). For colon tumors, the CNR of DLIR-H (5.4 ± 2.2) was 80% higher than that of ASIR-V 50% (3.0 ± 1.1, P < 0.001). Although ASIR-V 50% maintained a higher ERS than DLIR-H (median, 108.3 vs. 101.5 HU/mm; P < 0.001), the ERD remained highly stable across all algorithms (median, 2.5 mm; pairwise P > 0.05). Subjectively, DLIR-H received the highest scores for overall image quality and regional lymph node visualization (median, 5.0 and 4.5, respectively, vs. 3.0 for both in ASIR-V 50%; all P < 0.001). In 40-keV virtual monoenergetic CT of colon cancer, DLIR-H significantly improves objective and subjective image quality for tumors, vessels, and lymph nodes. While a minor objective edge-smoothing effect exists, DLIR-H provides an optimal balance between robust noise suppression and anatomical clarity. Although these findings suggest the potential to facilitate low-iodine spectral protocols, future diagnostic accuracy trials are required to confirm their true clinical impact.
- 6 April 2026
- Published by Springer Nature in Abdominal Radiology
- Vol. 51 (10) , 4820-4833
Deep learning reconstruction for 40-keV virtual monoenergetic CT of colon cancer: evaluation of image quality and edge sharpness
Virtual monoenergetic imaging (VMI) at 40 keV improves iodine attenuation in colon cancer CT but is constrained by severe image noise. Deep learning image reconstruction (DLIR) may address this limitation, but its effect on anatomical edge preservation across multiple targets requires investigation. To evaluate the impact of DLIR on objective and subjective image quality of 40-keV VMIs in colon adenocarcinoma, with emphasis on the trade-off between noise reduction and edge definition. In this completed retrospective study (patient enrollment window: May 2024 to February 2025), 60 patients (mean age, 62.8 years ± 15.1; 34 men) with confirmed colon adenocarcinoma underwent dual-energy CT using a low-iodine protocol (1.0 mL/kg). Portal venous phase data were reconstructed at 40 keV using adaptive statistical iterative reconstruction-V (ASIR-V) 50%, medium-strength DLIR (DLIR-M), and high-strength DLIR (DLIR-H). Objective metrics, including contrast-to-noise ratio (CNR), edge rise distance (ERD), and edge rise slope (ERS), were measured. Two radiologists independently scored five qualitative categories, including regional lymph node visualization, using a 5-point Likert scale. DLIR-H yielded the lowest image noise and highest CNR across all anatomical targets compared with DLIR-M and ASIR-V 50% (all P < 0.001). For colon tumors, the CNR of DLIR-H (5.4 ± 2.2) was 80% higher than that of ASIR-V 50% (3.0 ± 1.1, P < 0.001). Although ASIR-V 50% maintained a higher ERS than DLIR-H (median, 108.3 vs. 101.5 HU/mm; P < 0.001), the ERD remained highly stable across all algorithms (median, 2.5 mm; pairwise P > 0.05). Subjectively, DLIR-H received the highest scores for overall image quality and regional lymph node visualization (median, 5.0 and 4.5, respectively, vs. 3.0 for both in ASIR-V 50%; all P < 0.001). In 40-keV virtual monoenergetic CT of colon cancer, DLIR-H significantly improves objective and subjective image quality for tumors, vessels, and lymph nodes. While a minor objective edge-smoothing effect exists, DLIR-H provides an optimal balance between robust noise suppression and anatomical clarity. Although these findings suggest the potential to facilitate low-iodine spectral protocols, future diagnostic accuracy trials are required to confirm their true clinical impact.