RPS 101 - Frontiers in colorectal cancer imaging: from diagnosis to precision therapy

Author Block: Z. Guan, Y-S. Sun; Beijing/CN
Purpose: To prospectively validate an AI-optimized colon MRI protocol integrating L₁-regularized compressed sensing (ACS) and deep learning reconstruction (DeepRecon) against contrast-enhanced CT (CECT) for staging accuracy and image quality.
Methods or Background: This prospective single-center study (August 2022-April 2024) included 132 patients with pathologically confirmed colon cancer undergoing both 3.0T MRI (tri-oblique planes) and CECT within 1 week. Overall image quality, sharpness of peritoneum, sharpness of colonic wall, tumor conspicuity, lymph node conspicuity and artifacts were assessed using a modified 4-point Likert scale. Contrast-to-noise ratio (CNR) and edge rise distance (ERD) were obtained and reported as medians and IQRs. Surgical pathology correlation was performed in 33 cases.
Results or Findings: This study included 153 matched examinations from 132 participants (mean age, 58 years ± 11 [SD]). The AI-optimized MRI protocol demonstrated statistically superior performance compared to CECT in both qualitative and quantitative assessments. MRI showed higher median overall image quality scores (4 [IQR 3-4] vs. 3 [3-3]), improved CNR (5.5 [IQR 4.2-7.0] vs. 1.3 [IQR 0.7-2.2]), and reduced ERD (1.6 mm [IQR 1.2-2.8] vs. 3.7 mm [IQR 2.8-5.0]). Qualitative analysis demonstrated superior performance of MRI in peritoneal sharpness (75.8%), colonic wall visualization (100%), tumor conspicuity (92.1% preferred/equal), and lymph node assessment (86.9% preferred/equal). However, CECT provided better artifact control (45.1% preferred). MRI achieved higher T-staging accuracy (78.8% vs. 30.3%) and N-staging specificity (93.3% vs. 13.3%) compared to CECT.
Conclusion: AI-optimized MRI protocol significantly outperformed CECT in image quality and preliminary staging accuracy while eliminating radiation, demonstrating potential as a preoperative assessment tool.
Limitations: It was conducted at a single center.
Funding for this study: Project funded by National Natural Science Foundation of China (U24A20752, 82471958, 82271955), Beijing Natural Science Foundation (L242038), Beijing Hospitals Authority Clinical Medicine Development of Special Funding Support (ZLRK202522), Capital's Funds for Health Improvement and Research (2024-1-1022), Peking University Medicine Sailing Program for Young Scholars’ Scientific & Technological Innovation, the Fundamental Research Funds for the Central Universities (BMU2025YFJHPY038).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Institutional Review Board of Peking University Cancer Hospita;

Author Block: F. Wu, C. Huang; Guiyang City, Guizhou Province/CN
Purpose: To investigate the value of CT-derived extracellular volume fraction (fECV) in predicting microsatellite instability (MSI) status in colon cancer.
Methods or Background: The clinical and imaging data of 168 patients with colon cancer who underwent radical surgery were retrospectively analyzed. The imaging features of the lesions were extracted and fECV was calculated. According to the postoperative results, the patients were divided into MSI group ( 17 cases ) and MSS group ( 151 cases ). The binary logistic regression model was constructed for combined diagnosis, and the ROC curve was used to evaluate the diagnostic efficacy.
Results or Findings: The differences in tumor location, differentiation grade, and lymph node status between the MSI and MSS groups were statistically significant (P<0.05); both fECV and △tumor values in the MSI group were higher than those in the MSS group (P<0.05). The area under the curve (AUC) for predicting MSI status using fECV, tumor location, differentiation grade, and lymph node metastasis were 0.834, 0.661, 0.667, 0.623, and 0.684 respectively. The imaging model (fECV + tumor location), pathological model (lymph node metastasis + differentiation grade), Model 1 (tumor location + pathological model), and Model 2 (fECV + Model 1) showed AUC, sensitivity, and specificity for predicting MSI status as follows: 0.915, 76.47%, 90.07%; 0.770, 94.12%, 51.66%; 0.818, 70.59%, 77.48%; 0.951, 100.00%, 79.47%. DeLong test suggests Model 2 was superior to all single indicators and models (P<0.05), followed by the imaging model which outperformed all single indicator models and pathological combination models (all P<0.05).
Conclusion: fECV have high diagnostic value for MSI and can be used to predict MSI status.
Limitations: This study is a single-center retrospective study lacking external validation; due to MSI-positive cases constraints, detailed subgroup analysis could not be conducted.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Hospital Ethics Committee (Approval No. 2025 Review No. 20250004)

Author Block: L. Zhu¹, K. Nie², Z. Zhao¹, Z. Zhou¹, J. Gao¹; ¹Zhengzhou/CN, ²Shanghai/CN
Purpose: To determine the optimal quantum iterative reconstruction (QIR) level in abdominal photon-counting detector CT (PCD-CT) for preoperative TNM staging of colorectal cancer (CRC), focusing on image quality and diagnostic confidence.
Methods or Background: A total of 98 consecutive patients with biopsy-confirmed colorectal cancer who underwent enhanced abdominal PCD-CT for preoperative staging were enrolled between September 2024 and February 2025. Virtual monoenergetic images (VMI) at 60 keV in the portal venous phase were reconstructed without QIR (QIR-off) and with QIR levels 1–4 (slice thickness: 1 mm). Quantitative metrics (signal-to-noise ratio [SNR], contrast-to-noise ratio [CNR]) of primary tumors, regional lymph nodes, and liver metastases were measured. Two blinded radiologists independently evaluated qualitative metrics of the image quality and diagnostic confidence for CRC T-staging and liver metastases with a 5-point Likert scale.
Results or Findings: From QIR-off to QIR-4, image noise progressively decreased, with corresponding significant improvements in both SNR and CNR for colorectal tumors, regional lymph nodes, and 38 liver metastatic lesions (all P < 0.001). Both readers rated QIR-4 superior to other reconstructions across all qualitative metrics, including conspicuity and edge sharpness of the colorectal tumors and lymph nodes, peritumoral vessel clarity, conspicuity of liver metastatic lesions, and overall image quality (all P < 0.01). Furthermore, QIR-4 improved the diagnostic confidence for CRC T-staging and the small liver lesions with maximum diameter <10mm compared with other reconstructions (all P < 0.05).
Conclusion: High-strength level QIR-4 on abdominal PCD CT provided optimal image quality for preoperative TNM-staging evaluation of colorectal cancer and improved the diagnostic confidence of the radiologists.
Limitations: This was a single-center study with a relatively small sample size.
Funding for this study: The study received no funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the institutional review board (no. 2024-KY-1106-002).

Author Block: Y. A. Sliem, M. A. Shaaban; 6th of October City/EG
Purpose: To perform a meta-analysis of the diagnostic accuracy of CT colonography in detecting colorectal polyps ≥6 mm.
Methods or Background: Background: Colorectal cancer (CRC) remains a leading cause of cancer-related morbidity and mortality worldwide. Early detection of adenomatous polyps is critical for prevention. Conventional colonoscopy, although highly accurate, is invasive and associated with low patient compliance. CT colonography (CTC) offers a noninvasive alternative, but reported diagnostic performance varies.

Methods: Systematic search of PubMed, Embase, and Cochrane databases (2005–2025) for studies comparing CTC with conventional colonoscopy as reference. Inclusion criteria: sensitivity and specificity data for polyps ≥6 mm, ≥50 patients. Pooled sensitivity, specificity, and diagnostic odds ratio were calculated using a random-effects model. Heterogeneity was assessed with I² and publication bias was evaluated.
Results or Findings: Fifteen studies (n=5,100 patients) met inclusion criteria. Pooled sensitivity and specificity for polyps ≥6 mm were 92% (95% CI: 89–94%) and 88% (95% CI: 85–91%), respectively. The diagnostic odds ratio was 102 (95% CI: 80–128). Subgroup analysis indicated higher sensitivity in studies using 64-slice CT scanners compared to earlier technology. Patient compliance increased by 25% with noninvasive CTC.
Conclusion: CT colonography is a reliable noninvasive method for colorectal polyp detection, with diagnostic accuracy approaching conventional colonoscopy. Adoption may improve screening uptake and early CRC prevention.
Limitations: Motion artifacts
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: A. Hinterberger¹, J. Bohn¹, J. Betge², S. Grosu³, S. O. Schönberg², D. Nörenberg², J. N. Kather⁴, F. Grawe¹, K. Maier-Hein¹; ¹Heidelberg/DE, ²Mannheim/DE, ³Munich/DE, ⁴Dresden/DE
Purpose: Noninvasive colorectal cancer (CRC) screening represents a key opportunity to improve colonoscopy participation rates and reduce CRC mortality. This study explores the potential of the gut-liver-axis for predicting colorectal neoplasia through liver-derived radiomic features extracted from routine CT images.
Methods or Background: In this retrospective study, we analyzed data of 1,997 patients who underwent colonoscopy and abdominal CT. Patients either had no colorectal neoplasia (n=1,189) or colorectal neoplasia (ntotal=808; adenomas n=423, CRC n=385). Radiomic features were extracted from 3D liver segmentations using the Radiomics Processing ToolKit (RPTK), which performed feature extraction, filtering, and classification. The dataset was split into training (n=1,397) and test (n=600) cohorts. Five machine learning models were trained with 5-fold cross-validation on the 20 most informative features and ensembles.
Results or Findings: The radiomics-based XGBoost model achieved best validation performance with a test AUROC of 0.810 [95% CI: 0.767–0.837], clearly outperforming the best clinical-only model (test AUROC: 0.457 [95% CI: 0.411–0.506]). After Youden index-based threshold optimization, the final model reached a test sensitivity of 74.1% and specificity of 72.3% for predicting the presence of colorectal neoplasia. Subclassification between colorectal cancer and adenoma showed lower accuracy (test AUROC: 0.674 [95% CI: 0.606–0.741]).
Conclusion: Our findings establish proof-of-concept that liver-derived radiomics from routine abdominal CT can predict colorectal neoplasia. Beyond offering a pragmatic, widely accessible adjunct to CRC screening, this approach highlights the gut–liver axis as a novel biomarker source and sparks new mechanistic hypotheses for future translational research.
Limitations: This single-center, retrospective study was intended as a proof-of-concept to demonstrate technical feasibility. Validation in larger, prospective, and multicenter cohorts will be essential to confirm generalizability and clinical utility.
Funding for this study: No funding or grant support was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethikkommission II (2023-887-AF 11)

Author Block: J. Leonhardi, A-K. Höhn, M. Mehdorn, T. Denecke, H-J. Meyer; Leipzig/DE
Purpose: Texture analysis provides quantitative imaging markers and might improve tumor tissue characterization. Node-RADS classification is a classification system to improve lymph node description in oncological imaging. The present analysis investigated the diagnostic benefit of both methods to categorize and stage lymph nodes in patients with colon cancer.
Methods or Background: For Node-RADS analysis, 108 patients (n = 49 females, 45.3%, age 70.08 ± 14.34 years) were included. All patients were surgically resected and lymph nodes were histopathological analyzed. Lymph nodes were scored in accordance to the Node-RADS classification by two experienced radiologists. 85 patients were eligible for texture analysis of the lymph nodes (n = 39 females, 45.9%). Interreader variability was assessed with Cohen's kappa analysis, discrimination analysis was performed with Mann-Whitney-U test and diagnostic accuracy was assessed with ROC curve analysis.
Results or Findings: Out of 279 extracted CT texture features, 7 parameters independently showed statistically significant differences between lymph node metastasis to negative lymphnodes. A multivariate model based on n = 7 independent texture parameters reached an AUC of 0.79 (95% CI: 0.69-0.89) with sensitivity of 0.77 and specificity of 0.70. Node-RADS scores showed statistically significant differences between N-/N + stage (N- mean 1.89 ± 1.09 score versus N+ 2.93 ± 1.62 score, p = 0.001). ROC curve analysis showed an AUC of 0.68 for Node-RADS-based lymph node discrimination, with an optimal sensitivity of 0.62 and specificity of 0.71.
Conclusion: Texture analysis can improve the diagnostic accuracy for nodal CT staging in patients with colon cancer. Node-RADS scores showed only limited diagnostic accuracy in colon cancer with high interreader variability. This should question clinical translation.
Limitations: This was a retrospective study design with possible known inherent bias. Further validation studies are needed to confirm present results.
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: This retrospective, observational study involving human participants was performed in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. This retrospective study was approved by our institutional ethics review board (Ethics Committee of the University of Leipzig EK: 243-14-14072014), and the requirement for informed consent of patients was waived.

Author Block: M. Asano, Y. Noda, N. Kawai, T. Suto, T. Kaga, S. OMATA, T. Miyoshi, H. Kato, M. Matsuo; Gifu/JP
Purpose: To evaluate the feasibility of 40-keV virtual monoenergetic images (VMIs) derived from dual-energy CT for assessing extramural vascular invasion (EMVI) in patients with rectal cancer.
Methods or Background: Fifty-six patients with rectal cancer underwent both contrast-enhanced dual-energy CT and rectal MRI were retrospectively analyzed. The 40-keV VMIs at the portal venous phase were reconstructed for all patients. Two radiologists evaluated the EMVI scores on 40-keV VMIs. Another radiologist interpreted rectal MRI and created the reference standard. The diagnostic performance for detecting EMVI (EMVI scores of 3 and 4) in 40-keV VMIs was compared with rectal MRI findings as a reference standard using McNemar test.
Results or Findings: The sensitivity and specificity for detecting EMVI in 40-keV VMIs were 85.7% and 88.6%, respectively. When comparing with rectal MRI, the 40-keV VMIs showed no statistical difference in the sensitivity (P =.25) and specificity (P =.13). The κ value was 0.42 in 40-keV VMIs, indicating moderate interobserver agreement.
Conclusion: When compared with rectal MRI, the 40-keV VMIs had no statistical difference in the sensitivity and specificity for diagnosing EMVI in patients with rectal cancer.
Limitations: Small sample size, evaluated only EMVI of rectal cancer, not compared with the pathological results, and only used rapid-kilovoltage-switching DECT.
Funding for this study: This study received no funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Our institutional review board approved this retrospective study and waived written informed consent due to the retrospective nature of the study.

Author Block: N. Gündüz, M. Gezgin, B. Akgun, H. Seneldir, B. Atalay, O. Alimoglu, G. Erdem; Istanbul/TR
Purpose: Rectal cancer affects more than 700,000 people annually worldwide and accounts for 3.2% of cancer-related deaths.Regional lymph node(LN) status is a critical determinant in TNM staging, treatment planning, and prognosis.Magnetic resonance imaging(MRI) is the preferred modality for local staging, yet its diagnostic reliability for LN involvement remains debated. This study aimed to evaluate the diagnostic performance and reliability of MRI in assessing LN metastasis in rectal cancer.
Methods or Background: This retrospective study included patients with rectal cancer who underwent preoperative pelvic MRI and subsequent surgical resection between 2020 and 2023. Patients without either imaging or pathology data were excluded. Two abdominal radiologists(1–10 years of experience) independently reviewed MRIs using the 2016 ESGAR criteria.LN morphology(short-axis size, round shape, irregular border, heterogeneity) was assessed. Nodes ≥9 mm, 5–9 mm with ≥2 malignant features, or <5 mm with ≥3 features were classified as malignant; mucinous nodes were also considered malignant. In patients receiving neoadjuvant therapy, nodes >5 mm were classified malignant. MRI findings were compared with histopathology. Interobserver agreement was analyzed with kappa statistics.
Results or Findings: Of 68 patients, 42 were eligible (mean age 57.6 years; 24 male, 18 female).Histopathology confirmed LN metastases in 16 cases, all detected by MRI(sensitivity 100%). Of 26 LN-negative cases, MRI correctly identified 15 but overcalled 11, yielding specificity 57.7%. Overall diagnostic accuracy was 73.8%, with positive predictive value 59.3% and negative predictive value 100%. Interobserver agreement was excellent (κ=0.94).
Conclusion: MRI demonstrated high sensitivity but relatively low specificity in detecting LN involvement in rectal cancer, consistent with prior literature. Awareness of potential false positives is essential, and refinement of morphologic malignancy criteria may improve specificity.
Limitations: Single-center design, retrospective analysis, modest sample size, and inability to stratify histopathological LN subtypes may limit generalizability.
Funding for this study: No funding
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approved by Istanbul Medeniyet University Ethics Committee, 2023/0292

Author Block: Y. Shen¹, T. Jiang¹, X. Gong², W. Meng¹, Z. Wang¹; ¹Chengdu/CN, ²Cheng du/CN
Purpose: Total neoadjuvant therapy (TNT) has improved the pathological complete response (pCR) rate for locally advanced rectal cancer (LARC). The relationship between tumor response after the induction chemotherapy (INCT) and pCR has not been decided. We aimed to evaluate the association between the post-induction (PI) tumor response after INCT and the ultimate treatment response in LARCs undergoing TNT.
Methods or Background: From 2018 to 2022, consecutive patients diagnosed with LARC who received TNT before radical surgery and had MRI examinations at both PI and pre-operation (PO) were included. 360 LARC patients were enrolled for initial assessment, with 224 patients included in the final analysis. The primary outcome was the pCR rate. MR-TRG and tumor longitudinal length reduction rate (TLLR) were used to evaluate the tumor response to INCT at PI. The combination of MR-TRG and DWI was used to determine the presence of MRI complete response (mriCR) at PO. A univariate logistic regression model was used to establish associations between PI response and PO response, PI response and pathological response.
Results or Findings: The current study included 224 patients (median [IQR] age, 56 [49-65] years; 136 [60.7%] male). Compared with patients who did not achieve partial response (PR) at PI, patients who achieved PR at PI have higher PO mriCR rate (OR 3.788, 95%CI 1.740-8.202, p<0.001) and higher final pCR rate (OR 4.412, 95%CI 1.881-10.347, p<0.001).
Conclusion: Tumor response to INCT is related to the final pathological response in LARC patients who underwent TNT. Good responders to INCT have higher odds of reaching a mriCR or pCR, supporting mid-term tumor response evaluation after INCT in a response-guided treatment strategy.
Limitations: Survival outcomes were not included in the current study because the data were still unavailable.
Funding for this study: This work was supported by the Postdoctoral Fellowship Program of CPSF (No. GZB20250525), the National Natural Science Foundation of China (82103541, 82203474, 82203394, 82103918), Department of Science and Technology of Sichuan Province (No. 2022YFS0162, 2021YFS0025, 2022YFS0209), Sichuan Province Natural Science Foundation Project (2022NSFSC0840, 2023NSFSC1834), 1·3·5 Project for Disciplines of Excellence-Clinical Research Incubation Project, and West China Hospital, Sichuan University (No. 22HXFH001, 2019HXFH031).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study received institutional review board approval and written informed consent was obtained from all participants.

Author Block: G. Gorini¹, L. Sali¹, B. Mallardi¹, C. Campanino¹, E. Betti¹, F. Battisti¹, S. Stoffel², P. Mantellini¹; ¹Florence/IT, ²Basel/CH
Purpose: To evaluate if offering behavioural economics messages or alternative tests improve participation of non-respondents in immunochemical faecal test (FIT) based colorectal cancer (CRC) screening.
Methods or Background: 20,225 non-respondents to FIT (aged 50-70 years) in a CRC screening program were randomized into six groups and invited by mail: 1) to FIT with standard letter (SL); 2) to FIT with SL plus feedback message (F); 3) to FIT with SL plus social norm message (MN); 4) to FIT with SL and combination of F+MN; 5) to sigmoidoscopy ; 6) to CT colonography. The primary outcome measure was the participation rate within 90 days from invitation. The trial was registered: ISRCTN11841256.
Results or Findings: Overall 1,106 non-respondents to FIT partecipated. Screening participation rates were 5.7% in SL, 7.4% in F , 6.7% in MN, 6.6% in F+MN, 2.0% in sigmoidoscopy, 4.1% in CT colonography groups. F group invitees were more likely to participate as compared to SL control group (aOR=1.32; 95%CI: 1.10-1.57), to sigmoidoscopy group (aOR=0.32; 95%CI: 0.24–0.42) and to CT colonography group (aOR=0.67, 95%CI:0.52–0.86).
Conclusion: Behavioural economics interventions can increase participation in FIT-based CRC screening programs more than offering different screening tests. In non-respondents to FIT CT colonography recorded a two-fold participation to screening as compared to sigmoidscopy.
Limitations: Detection rates of colorectal neoplasia were not evaluated in presente study.
Funding for this study: Tuscany Region (CUP F18D20000120002, 2018)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Comitato Etico Regione Toscana- Area Vasta Centro (CEAVC): ID: 16875_spe, approved 19/01/21

Author Block: M. Kabiri, M. Nejati, C. Booz, I. Yel, K. Eichler, T. Vogl; Frankfurt am Main/DE
Purpose: The differentiation of hepatic cysts and colorectal cancer liver metastases (CRLM)
is critical for accurate diagnosis and treatment planning. This study evaluates the diagnostic
accuracy of Hounsfield Units (HU) in Virtual Non-Contrast (VNC) imaging compared to True
Non-Contrast (TNC) imaging for distinguishing these lesions.
Methods or Background: A retrospective analysis was conducted on 120 patients who underwent Dual-Energy
Computed Tomography (DECT) scans between January 2020 and December 2022. Patients with
confirmed hepatic cysts or CRLM were included. HU values were measured in regions of interest
(ROIs) within the lesions. Statistical analyses included paired t-tests and ROC curve analysis to
evaluate diagnostic performance.
Results or Findings: The TNC method achieved an AUC of 0.864 (sensitivity: 70.51%, specificity: 82.26%),
while the VNC method achieved an AUC of 0.861 (sensitivity: 66.88%, specificity: 84.62%). No
significant difference was observed between the two methods (p = 0.850). VNC imaging
demonstrated slightly higher mean HU values for both cysts and metastases compared to TNC
imaging.
Conclusion: VNC imaging offers comparable diagnostic accuracy to TNC imaging for
differentiating hepatic cysts and CRLM, with the added benefit of reduced radiation exposure.
These findings support the integration of VNC imaging into routine clinical practice.
Limitations: The limitations of this study are minimal and unlikely to impact the validity of the findings.
Funding for this study: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The institutional ethic review board approved the present retrospective single-center study with a waiver for written informed consent.

Author Block: A. Ammirabile¹, F. Mastroleo², R. De Martino¹, M. Waddle², A. Laghi¹, L. Viganò¹; ¹Milan/IT, ²Rochester, MN/US
Purpose: Chemotherapy-Associated Steatohepatitis (CASH) increases the risk of postoperative liver failure and 90-day mortality after hepatic resection. Although hepatomegaly is common, non-invasive diagnosis remains challenging: liver biopsy samples a limited parenchymal portion, and imaging cannot reliably distinguish simple steatosis from CASH. A radiomic-based virtual biopsy (VB) approach has shown promising results. This study evaluated CT-based radiomics for preoperative CASH prediction, comparing VB and whole liver parenchyma.
Methods or Background: This retrospective analysis included patients from the public Colorectal-Liver-Metastases dataset (internal) and from Costa et al. (2021) (external). Inclusion criteria were: portal-phase CT (thickness ≤5mm) after preoperative chemotherapy for colorectal liver metastases and pathological CASH assessment. IBSI-compliant radiomic features were extracted with PyRadiomics from total liver parenchyma (Liver-VOI) and VB (manually segmented 2 mL cylinder of non-tumoral parenchyma on 3D Slicer). Liver volume was estimated using three orthogonal diameters. Four XGBoost predictive models were developed (70/30 split, 5-fold cross-validation): clinical; clinical+VB-radiomics (ViB); clinical+VB-radiomics+liver volume (ViBE+); clinical+Liver-VOI radiomics (WholeLiver). ViBE+ underwent external validation. Performance was statistically compared by AUC, with SHAP analysis for clinical interpretability.
Results or Findings: 180 patients were included: 105 in the internal (63 males, mean age 58.9 years, 38% NASH+) and 75 in the external (44 males, mean age 62.4 years, 19% NASH+) cohorts. At internal validation, the clinical model had the lowest performance (AUC=0.713). WholeLiver and ViBE+ achieved the best prediction (0.857 vs. 0.831, p=0.814), while ViB showed AUC=0.724. Liver volume metrics emerged as key predictors in SHAP analysis. VIBE+ achieved AUC=0.745 at external validation.
Conclusion: VB provides a reproducible, non-invasive method for CASH prediction. Combined with liver size metrics, it achieves performance comparable to whole-liver analysis and supports clinically applicable preoperative risk stratification.
Limitations: Retrospective design; Modest sample size; Chemotherapy-only etiology of steatohepatitis.
Funding for this study: AIRC grant #2019−23822
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The local review board approved the study protocol (83/20). Because of the retrospective nature of the study, the need for specific informed consents was waived.