RPS 1701 - Liver tumour detection and characterisation

Author Block: Y. A. Sliem, M. A. Shaaban; 6th of October City/EG
Purpose: Perform a meta-analysis evaluating the diagnostic accuracy of AI-assisted multiphasic CT for detection and characterization of liver lesions and to compare AI performance with conventional interpretation.
Methods or Background: Background:Early and accurate detection of hepatic lesions,including primary hepatocellular carcinoma and metastases,is critical for patient prognosis and treatment planning.Multiphasic CT is a standard imaging modality; however;radiologist interpretations can be subjective and time-intensive,with potential for missed small or subtle lesions. Artificial intelligence(AI),particularly convolutional neural networks (CNNs), has been increasingly applied to medical imaging for automated lesion detection and characterization.

Methods: A systematic literature search was conducted in PubMed, Embase, and Scopus from January 2015 to June 2025 for studies evaluating AI-assisted detection of liver lesions on multiphasic CT. Inclusion criteria were original studies reporting sensitivity and specificity, comparison with radiologists, and histopathology or follow-up as the reference standard.Data were extracted independently by two reviewers.Pooled sensitivity,specificity,positive predictive value(PPV), negative predictive value(NPV),and diagnostic odds ratios(DOR)were calculated using random-effects model.Heterogeneity was assessed with I² statistics,and publication bias was evaluated using funnel plots.
Results or Findings: Thirteen studies(n=3,400 patients)met inclusion criteria. Pooled sensitivity and specificity of AI-assisted detection were 93%(95% CI: 90–95%)and 91%(95% CI: 88–94%),respectively.PPV and NPV were 90% and 94%.The diagnostic odds ratio was 145(95% CI: 110–180).AI outperformed radiologists in small lesion detection(<1 cm)with an odds ratio of 2.1(p<0.01).Interobserver agreement increased with AI assistance(Cohen’s kappa from 0.73 to 0.88).Subgroup analyses showed consistent performance across primary and metastatic lesions and various CT protocols. Moderate heterogeneity was observed(I²=45%).
Conclusion: AI-assisted multiphasic CT demonstrates high diagnostic accuracy for liver lesion detection and can enhance radiologist performance, particularly for small lesions. Integration of AI into clinical workflow may improve early diagnosis,reduce variability,and optimize patient management.
Limitations: Future research should focus on prospective multicenter validation and cost-effectiveness analyses.
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: C. Van De Braak, F. Willemssen, R. De Man, A. Van Der Lugt, D. Bos, R. S. Dwarkasing, .. SMS Consortium; Rotterdam/NL
Purpose: Current guidelines recommend biannual ultrasound (US) screening for high-risk patients for developing hepatocellular carcinoma (HCC), however it has limited sensitivity for early-stage detection of 47%. Our aim is to validate a Short MRI Surveillance (SMS)-protocol for HCC screening, and compare this to US in a prospective multicentre study. Here, we present our preliminary results.
Methods or Background: From November 2023, patients included from the current HCC surveillance programme underwent paired US-MRI screening. The abbreviated MRI (aMRI) was performed on 1.5/3.0-T systems from different vendors, using dedicated 8-16 channel range body coil. The SMS-protocol consisted of T1W in-out phase, T2W with fat saturation, and DWI. One radiologist evaluated both US and aMRI images, while two additional radiologists solely evaluated aMRI images. Statistical analyses included inter-rater reliability- and sensitivity/specificity analysis.
Results or Findings: For final analyses, 82 participants were included. The most common indication was hepatitis B (53.7%), and cirrhosis (32.9%). Based on US, a total of 20 lesions were found in 11 patients, whereas aMRI detected 132 lesions in 29 patients. Most findings were benign, except for two lesions that were noted on SMS, but undetected on US, and proved to be ≥ LIRADS-3 on subsequent contrast-enhanced MRI, i.e. the golden standard. There was substantial agreement between the readers (Gwet’s AC1: 0.70). There was no difference in sensitivity/specificity between US and SMS for detection of suspicious lesions (0.50/0.99, and 0.50/1.00, respectively).
Conclusion: Our preliminary results provide a proof of principle that aMRI can be a potential screening modality for HCC surveillance. At this timepoint, with the low number of events, no difference between the sensitivity/specificity of US and SMS could be established. Ongoing enrolment and follow-up are expected to provide more insight.
Limitations: Preliminary findings due to ongoing inclusions
Funding for this study: Funding was provided by a grant of the Dutch Cancer Society (KWF, grant number: 2021-2/13803)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the institutional review board, and written informed consent was obtained from all participants.

Author Block: L. Jia¹, J. Li¹, Y. Li², S. N. Li¹, J. Lei¹; ¹Lanzhou/CN, ²Beijing/CN
Purpose: While ultrasound has low sensitivity for hepatocellular carcinoma(HCC) screening in high-risk patients, this is particularly pronounced in the growing population with non-viral liver disease (e.g., MASH/NAFLD). This meta-analysis aimed to evaluate the diagnostic accuracy and cost-effectiveness of abbreviated MRI (AMRI) specifically in this subpopulation, and to compare the performance of different AMRI protocols.
Methods or Background: PubMed, Embase, Cochrane Library, and Web of Science databases were searched for studies published from inception to March 2025. A pre-specified subgroup analysis was performed for studies focusing on patients with non-viral etiologies of liver disease. Additionally, we directly compared the diagnostic accuracy of hepatobiliary phase AMRI (HBP-AMRI) versus non-contrast AMRI (NC-AMRI) protocols. qualitatively compared the cost-effectiveness of AMRI protocol with ultrasound in HCC screening.
Results or Findings: 17 studies (including 3290 high-risk patients) characterized the accuracy of AMRI for surveillance of HCC. The pooled sensitivity and specificity were 88% (95% CI: 84,91) and 95% (95% CI: 92,96). Six studies directly compared diagnostic accuracy of AMRI and ultrasound monitoring end-to-end, the sensitivity of ultrasound was significantly lower at 39% (95% CI: 23,57) than the sensitivity of AMRI at 82% (95% CI: 0.74,0.88, p < 0.001). Hepatobiliary phase contrast-enhanced AMRI (HBP-AMRI) had higher sensitivity than non-contrast AMRI (NC-AMRI) (91% [95% CI: 87, 95] vs 86% [95% CI: 82, 90], p 0.001),also higher than contrast enhanced AMRI (CE-AMRI) (91% [95% CI: 87, 95] vs 90% [95% CI: 86, 95], p 0.001). Most studies found semi-annual abbreviated MRI cost-saving versus ultrasound.
Conclusion: Abbreviated MRI demonstrates high performance in HCC screening for patients with both viral and non-viral liver disease and is cost-effective or cost-saving.
Limitations: No
Funding for this study: No
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: L. Asmundo¹, N. Mercaldo², A. J. Herold³, F. Vernuccio⁴, D. V. Sahani², M. Renzulli⁵, G. Brancatelli⁴, A. Vanzulli¹, O. A. Catalano²; ¹Milan/IT, ²Boston, MA/US, ³Vienna/AT, ⁴Palermo/IT, ⁵Bologna/IT
Purpose: To assess the 1-year natural history of LI-RADS 3 observations on contrast-enhanced MRI in cirrhotic patients across multiple international centers, and to identify clinical and imaging predictors of progression using multivariable and machine learning models.
Methods or Background: This retrospective study included 347 cirrhotic patients with 540 LI-RADS 3 observations from six centers across three countries, each with 12-month MRI follow-up. Observations were reassessed using LI-RADS v2018 criteria. Generalized linear mixed-effects models and machine learning (LASSO, random forest) evaluated predictors of progression. Area under the curve (AUC) analysis assessed the predictive performance of clinical and imaging variables.
Results or Findings: Within one year, 27% of LI-RADS 3 observations progressed: 13% to LI-RADS 4 and 14% to LI-RADS 5. Independent predictors of progression included lesion size (OR: 1.12, 95% CI: 1.01–1.24), Child-Pugh Class C (OR: 8.36, 95% CI: 1.01–69.27), and alcohol-related liver disease (OR: 0.24, 95% CI: 0.06–0.94). Enhancing capsule and untreated HCV were significant in univariable analysis. Imaging features improved predictive accuracy, increasing AUC from 0.65 to 0.72 (p = 0.01). A lesion size cut-off of 9.5 mm was associated with increased progression risk.
Conclusion: One in four LI-RADS 3 observations progress within one year. Lesion size, liver function, and etiology are key predictors. Integration of imaging features enhances risk stratification and supports more personalized follow-up strategies for indeterminate liver lesions.
Limitations: The potential selection bias and variability in imaging protocols across centers. Follow-up was limited to contrast-enhanced MRI within 12 months, excluding some patients. Reliance on imaging rather than histopathology may have caused misclassification. Ancillary features were applied in LI-RADS categorization but not independently analyzed to reduce inter-reader variability.
Funding for this study: This research received no external funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Institutional review board approval was obtained and informed consent was waived due to the retrospective design. Patient-identifiable information remained secure and encrypted on site servers.

Author Block: A. Aslam¹, J. Cowen², T. Parry¹, E. J. Wigmore³, N. Abimanue², A. Gangi-Burton⁴, A. Aravinthan⁴, J. Franklin⁵; ¹London/UK, ²Portsmouth/UK, ³Manchester/UK, ⁴Nottingham/UK, ⁵Bournemouth/UK
Purpose: Six-monthly surveillance for patients with hepatitis B or cirrhosis aims to detect hepatocellular carcinoma (HCC) at an earlier, and more treatable, stage. This UK-based multi-centre service evaluation described the proportion, demographics, and staging of HCC patients diagnosed through surveillance, symptomatically, or incidentally.
Methods or Background: This retrospective service evaluation was conducted by the Radiology Academic Network for Trainees (RADIANT). Consecutive patients with a first diagnosis of HCC from 1st January 2020 across 14 centres were included. Data were collected from electronic patient records and Radiology Information Systems, and presented using descriptive statistics.
Results or Findings: A total of 723 patients were included (79% [570/723] male, median 71 years [IQR 64-77]). Only 28% (201/723) were diagnosed with HCC through surveillance, whilst 48% (346/723) were diagnosed through symptomatic presentation and 24% (176/723) incidentally. Although 51 % (369/723) of patients had known cirrhosis pre-diagnosis, only 36% (257/723) had been enrolled in a local surveillance programme. A further 20% of patients (143/723) were confirmed to have cirrhosis at diagnosis. Surveillance-diagnosed patients were younger (median 67 years [IQR 61-74]) compared to symptomatic (median 72 years [IQR 65-79]) or incidental (median 73 years [IQR 65-79]) groups, and also had a greater proportion of BCLC 0/A HCC (55%, 110/201), compared to symptomatic (26%, 91/346) or incidental (35%, 62/176) groups. Patients diagnosed symptomatically had highest proportions of ascites, vascular invasion, metastases, and lowest 3-, 6-, and 12-month survival.
Conclusion: In the UK, most patients with HCC are diagnosed outside of surveillance. Surveillance-diagnosed patients are identified at earlier stages of disease, with lower mortality rate, reinforcing the need to improve how we identify high-risk patients, enrol and maintain these patients in surveillance, and deliver surveillance itself.
Limitations: Retrospective design.
Funding for this study: N/A
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: E. J. Wigmore¹, N. Abimanue², T. Parry³, A. Aslam³, J. Cowen², A. Gangi-Burton⁴, A. Aravinthan⁴, J. Franklin⁵; ¹Manchester/UK, ²Portsmouth/UK, ³London/UK, ⁴Nottingham/UK, ⁵Bournemouth/UK
Purpose: Hepatocellular carcinoma (HCC) prognosis is suboptimal, typically due to late-stage diagnosis. Although six-monthly ultrasound surveillance (USS) is recommended for high-risk patients, absence of a centralised national programme results in inconsistent implementation. This United Kingdom (UK) multicentre service evaluation assessed provider adherence and patient compliance with HCC USS.
Methods or Background: The Radiology Academic Network for Trainees (RADIANT) retrospectively identified consecutive adult patients receiving ≥1 HCC USS in 2021 using the Radiology Information System of 17 participating centres. Records from 2022 were reviewed to assess adherence to six-monthly USS. Provider adherence was defined by scan requests; patient compliance by attendance. Data were presented using descriptive statistics and risk differences with chi-squared confidence intervals.
Results or Findings: A total of 1713 patients (58% [993/1713] male; median age 62 years [IQR 54-70]) were included, with 2285 HCC USS requested, of which 88% (2018/2285) were attended. Provider full (2/2 USS), partial (1/2 USS) and non-adherence was 49% (839/1713), 35% (607/1713) and 16% (267/1713) respectively. Of the 84% (1446) of patients for whom ≥1 USS had been requested, full, partial and non-compliance rates were 83% (1199/1446), 11% (160/1446), and 6% (87/1446) respectively. The absolute risk of none relative to full compliance for patients with alcohol history relative to those without was 3% (95% CI 0, 6) higher. For patients with ≥2 comorbidities relative to those without, the absolute risk of none relative to full compliance was 2% (95% CI -1, 4) lower.
Conclusion: UK surveillance delivery is suboptimal, due to deficits in provider adherence and patient compliance. National investment in HCC surveillance infrastructure, including tracking and recall systems, is needed, as well as targeted interventions to promote patient compliance.
Limitations: Retrospective. Patients eligible for USS but not scanned in 2021 were not captured.
Funding for this study: N/A
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: L. Jia, J. Lei; Lanzhou/CN
Purpose: P53-mutated hepatocellular carcinoma (HCC) is a highly aggressive subtype with profound implications for prognosis and treatment planning. However, non-invasive preoperative identification remains challenging. The purpose of this study was to assess the diagnostic utility of enhanced CT imaging characteristics for P53-mutated HCC and to elucidate their imaging characteristics.
Methods or Background: Patients with surgically resected HCC in a university hospital between January 2020 and December 2023 were retrospectively evaluated. Two radiologists independently evaluated CT imaging features and assessed inter-reader agreement.The Mann-Whitney test, 2 test, and logistic regression analysis were used to compare the enhanced CT characteristics, clinical features, and pathologic results. Using area under the curve, sensitivity and specificity to assess model performance.
Results or Findings: A total of 194 patients (mean age, 56 ± 10 years; 151 men) were included. P53 mutations were significantly associated with several aggressive pathological features (p-values ranging from <0.001 to 0.016). On enhanced CT, the presence of intratumoral arteries (p < 0.001) and an arterial phase hypovascular component (p < 0.001) were strong predictors of mutation. Multivariate analysis confirmed these as independent predictors, with high odds ratios of 9.4 (95% CI: 3.3, 26.8) and 10.6 (95% CI: 4.1, 27.6), respectively. The logistic regression model integrating these two features demonstrated excellent diagnostic performance, with an AUC of 0.82 (95% CI: 0.76, 0.89), a sensitivity of 71%, and a specificity of 79%.
Conclusion: Our study identifies two specific enhanced CT features—intratumoral arteries and an arterial phase hypovascular component—that are independently associated with P53 mutations in HCC. The model enables noninvasive, preoperative detection of this aggressive HCC subtype to guide risk stratification and personalize treatment.
Limitations: None
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the institutional review boards of all participating centers, and the requirement for obtaining informed consent from patients was waived.

Author Block: D. Choi¹, H. Rhee¹, Y. N. Park¹, H. G. Woo², T. Chung¹, J-h. Choi², J. E. Yoo¹; ¹Seoul/KR, ²Suwon/KR
Purpose: To identify gadoxetic acid-enhanced MRI findings associated with transcriptomic subtypes of hepatocellular carcinoma (HCC).
Methods or Background: We retrospectively enrolled treatment-naïve patients with pathologically confirmed HCC who underwent surgery between January 2008 and December 2019. Eligible patients had preoperative gadoxetic acid–enhanced MRI within 3 months of surgery and available RNA sequencing data. In cases with multiple tumors, only the largest lesion was analyzed. Transcriptomic subtypes were classified according to Montironi, Hoshida, and Chiang frameworks. Two radiologists independently evaluated major, ancillary, and LR-M LI-RADS features, as well as prognostic imaging findings on gadoxetic acid–enhanced MRI.
Results or Findings: This study included 89 HCCs from 89 patients (median age, 55 years; interquartile range, 49–64 years; 68 men). By Montironi’s classification, 36 tumors (40%) were inflamed and 53 (60%) non-inflamed. Hoshida’s system identified S1 (n=26, 29%), S2 (n=16, 18%), S3 (n=38, 43%), and unclassified (n=9, 10%) tumors. Chiang’s framework grouped them into CTNNB1 (n=17, 19%), proliferation (n=32, 36%), interferon (n=8, 9%), polysomy 7 (n=15, 17%), unannotated (n=8, 9%), and unclassified (n=9, 10%). In the proliferation subtype, marked HBP hypointensity was positively correlated (OR 4.214, P=0.005). For Hoshida S1, peritumoral HBP hypointensity (OR 5.715, P=0.028) and transitional-phase targetoid appearance (OR 18.118, P=0.037) were positive correlates, while capsule presence was negative (OR 0.276, P=0.037). In CTNNB1, marked HBP hypointensity was negatively correlated (OR 0.137, P =0.001) and intratumoral fat was positive (OR 3.980, P=0.040). For the inflamed subtype, capsule presence (OR 0.243, P=0.007) and substantial necrosis (OR 0.062, P=0.019) were negative correlates, with no other significant associations.
Conclusion: Ancillary, LR-M features, and prognostic imaging features on gadoxetic acid-enhanced MRI were significantly associated with established transcriptomic subtypes of HCC.
Limitations: Limitations include the single-center, retrospective design, lack of validation, and small sample size.
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 Institutional Review Board of Severance Hospital.

Author Block: P. Hou; Zhengzhou/CN
Purpose: To prospectively compare half-dose ultrahigh-resolution (UHR) photon-counting CT (PCCT) with full-dose standard-resolution energy-integrating detector CT (EID-CT) for detecting small liver metastases.
Methods or Background: Fifty patients with confirmed or suspected liver metastases underwent paired contrast-enhanced abdominal scans: standard-dose EID-CT (SD-SR) followed by half-dose UHR-PCCT (120×0.2-mm collimation), as LD-UHR. Four blinded radiologists independently analyzed all scans, marking lesions, rating confidence (5-point scale), and assessing image quality. Lesion detection sensitivity, inter-/intrareader reproducibility (intraclass correlation coefficient [ICC]), and measurement accuracy were compared using McNemar tests, paired t test, and mixed-effects models.
Results or Findings: Consensus sensitivity for all lesions was comparable between PCCT and EID-CT (87.3% [144/165] vs 91.1% [163/179], P=0.13). PCCT demonstrated higher reader confidence (mean score: 4.5±0.3 vs 4.2±0.7, P<0.05) and superior intrareader reproducibility (ICC>0.90 vs. 0.75-0.85 for EID-CT). Notably, for lesions <0.5 cm, sensitivity showed a modest reduction with PCCT (68.2% [45/66] vs 75.8% [50/66], P<0.001), though still maintaining diagnostic acceptability for subcentimeter lesions .Radiation dose was reduced by 52% with PCCT (3.1 vs 6.5 mSv, P<0.001). Multivariable analysis confirmed PCCT’s independent association with improved measurement precision (odds ratio:1.42, P=0.02).
Conclusion: Half-dose UHR-PCCT achieves diagnostic sensitivity equivalent to full-dose EID-CT for liver metastases >0.5 cm while reducing radiation exposure by half, offering a viable low-dose strategy for longitudinal oncologic surveillance without compromising detection efficacy for clinically actionable lesions.
Limitations: Not applicable.
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 ethics committee notification can be found under the number (No. 2022-KY-0752-001).