European Journal of Radiology最新文献

{"title":"Deep transfer learning radiomics combined with explainable machine learning for predicting malignancy risk in parotid gland tumors based on ultrasound.","authors":"Wei Wei, Wang Zhou, Chen Chen, Tian Jiang, Fei Xia, HuiJun Feng, Tianjun Wei, Dong Xu, Chaoxue Zhang","doi":"10.1016/j.ejrad.2026.112826","DOIUrl":"10.1016/j.ejrad.2026.112826","url":null,"abstract":"<p><strong>Objective: </strong>This study aimed to develop and validate an ultrasound (US)-based deep transfer learning radiomics model, integrated with explainable machine learning, for the preoperative malignant risk prediction of parotid gland tumors (PGTs).</p><p><strong>Methods: </strong>Data from 1,191 patients were retrospectively collected from three medical centers, and postoperative histopathological examination was used as the reference standard. Radiomics features and deep transfer learning features (ResNet50, Inception_V3, Vgg19) were extracted from the US images. Key predictive variables were selected using principal component analysis (PCA) and the least absolute shrinkage and selection operator (LASSO). Six classifiers-decision tree, gradient boosting machine, k-nearest neighbors, logistic regression, naïve Bayes, and random forest-were employed to construct models based on five feature sets: Clinical model, radiomics (Rad) model, deep transfer learning radiomics (DTLR) model, combined deep transfer learning and radiomics (DTLR-Rad) model, and a comprehensive combined model (CM _{Clinical + DTLR-Rad}). Model performance was evaluated using the area under the curve (AUC). Feature importance was interpreted using SHapley Additive exPlanations (SHAP). A web application for real-time, personalized risk prediction was developed.</p><p><strong>Results: </strong>In external test sets 1 and 2, the CM _{Clinical + DTLR-Rad} model based on the random forest classifier achieved the highest AUCs among the evaluated models, with 0.922 (95% CI: 0.890-0.954) and 0.959 (95% CI: 0.932-0.985), respectively. The integrated model outperformed the clinical-only and single-modality models in both external test sets. SHAP visualizations demonstrated the contribution of individual features. The web application provided both prediction probabilities and feature-level interpretability.</p><p><strong>Conclusion: </strong>The CM _{Clinical + DTLR-Rad} model demonstrated good predictive performance. The integration of interpretable machine learning and a web-based application may enhance preoperative risk stratification for PGTs and support clinical decision-making.</p>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":" ","pages":"112826"},"PeriodicalIF":3.3,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147698039","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Corrigendum to “Short occipital circulation time derived from quantitative digital subtraction angiography is associated with headache risk in patients with unruptured brain arteriovenous malformations” [Eur. J. Radiol. 192 (2025) 112402]","authors":"Yong-Sin Hu ,&nbsp;Jr-Wei Wu ,&nbsp;Huai-Che Yang ,&nbsp;Hsiu-Mei Wu ,&nbsp;Cheng-Chia Lee ,&nbsp;Feng-Chi Chang ,&nbsp;Kang-Du Liu ,&nbsp;Chung-Jung Lin","doi":"10.1016/j.ejrad.2026.112719","DOIUrl":"10.1016/j.ejrad.2026.112719","url":null,"abstract":"","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112719"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146161988","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The value of a multimodal ultrasound radiomics-based nomogram in predicting central lymph node metastasis of papillary thyroid microcarcinoma","authors":"Minhao Lin ,&nbsp;Xiaohong Xu ,&nbsp;Jianling Peng ,&nbsp;Qiuxia Huang ,&nbsp;Jiajian Wu ,&nbsp;Lijuan Liu","doi":"10.1016/j.ejrad.2026.112725","DOIUrl":"10.1016/j.ejrad.2026.112725","url":null,"abstract":"<div><h3>Objectives</h3><div>This study aimed to develop a multimodal ultrasound-based nomogram integrating radiomic features from grayscale ultrasound (GSUS) and contrast-enhanced ultrasound (CEUS) with clinical-ultrasound factors for the noninvasive prediction of central lymph node metastasis (CLNM) in patients with papillary thyroid microcarcinoma (PTMC), and to evaluate its predictive performance to support clinical decision-making.</div></div><div><h3>Methods</h3><div>This retrospective cohort study included 449 pathologically confirmed PTMC patients from June 2023 to December 2024, randomly divided into training (n = 314) and validation (n = 135) cohorts. Radiomic features were extracted using PyRadiomics software, and feature selection was performed through Spearman correlation analysis and LASSO regression. Multivariate regression analysis identified independent clinical risk factors for CLNM. A multimodal ultrasound combined model was then developed, serving as the basis for the nomogram. The model’s discriminative ability, calibration performance, and clinical utility were evaluated using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA).</div></div><div><h3>Results</h3><div>Multivariate analysis identified male sex, age &lt; 40 years, and capsular invasion as independent risk factors for CLNM. Single-modal models (Clinical, GSUS, CEUS) achieved AUCs ranging from 0.654 to 0.787 in the validation cohort. The Combined model integrating these features significantly outperformed all single-modal ones, with AUCs of 0.925 and 0.885 in the training and validation cohorts. Calibration curves and DCA confirmed its good fit and high clinical net benefit.</div></div><div><h3>Conclusion</h3><div>We successfully developed and validated a nomogram model based on multimodal ultrasound features for accurately predicting CLNM risk in PTMC patients, highlighting the value of radiomics in clinical risk assessment.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112725"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146212681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Interpretable machine learning based on intratumoral and peritumoral ultrasound radiomics for predicting central lymph node metastasis in papillary thyroid carcinoma","authors":"Wanting Yang ,&nbsp;Xuejiao Su ,&nbsp;Can Yue,&nbsp;Weizheng Chen,&nbsp;Yang Chen,&nbsp;Yan Luo,&nbsp;Buyun Ma","doi":"10.1016/j.ejrad.2026.112727","DOIUrl":"10.1016/j.ejrad.2026.112727","url":null,"abstract":"<div><h3>Objectives</h3><div>This retrospective and single-center study aimed to develop machine learning (ML) model integrating clinical features, ultrasound (US) features, and radiomics signatures extracted from both intratumoral and peritumoral regions to predict central lymph node metastasis (CLNM) in papillary thyroid carcinoma (PTC). The SHapley Additive exPlanations (SHAP) method was applied to visualize the prediction process and enhance clinical interpretability.</div></div><div><h3>Materials and methods</h3><div>A total of 879 patients with PTC who underwent preoperative US examination between January 2023 and January 2024 were retrospectively analyzed. Patients were randomly divided into training (n = 615) and test (n = 264) sets. Radiomics signatures were extracted from intratumoral regions and peritumoral regions extending 3 mm and 5 mm beyond the tumor margin. After feature selection, Radscore were computed. Five ML models incorporating clinical features, US features and Radscore were developed. Model performance was evaluated using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), calibration curves, and the Hosmer–Lemeshow test. SHAP was used to explain ML model predictions.</div></div><div><h3>Results</h3><div>CLNM occurred in approximately 52% of PTC. Patients with CLNM were younger and more often male (p &lt; 0.001). Multifocal tumors, extrathyroidal extension, and suspicious lymph nodes on US were also associated with higher CLNM risk (p &lt; 0.05). The Radscore derived from intratumoral and peritumoral regions were significantly different between patients with and without CLNM (p &lt; 0.05). Combined ML models outperformed those based on clinical and US features (p &lt; 0.05). The best performing model (XGB) achieved an AUC of 0.868 (sensitivity = 0.777, specificity = 0.803 and accuracy = 0.749) in the training set and an AUC of 0.787 (sensitivity = 0.704, specificity = 0.695 and accuracy = 0.713) in the test set. The XGB model demonstrated superior clinical utility and well-calibrated for CLNM prediction. SHAP analysis identified the Radscore from the combination of intratumoral and 3-mm peritumoral regions as the most CLNM predictor and provided patient-level interpretability.</div></div><div><h3>Conclusions</h3><div>Intratumoral and peritumoral radiomics features based on US show potential for predicting CLNM in PTC. The integration of SHAP analysis enhances model transparency and may support individualized treatment decision-making.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112727"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146212672","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Differentiating large-duct pancreatic ductal adenocarcinoma from malignant intraductal papillary mucinous neoplasm: MRI characteristics and diagnostic implications","authors":"Se Jin Choi ,&nbsp;Dong Wook Kim ,&nbsp;Byung-Kwan Jeong ,&nbsp;Seung-Mo Hong ,&nbsp;Jae Ho Byun ,&nbsp;Seung Soo Lee ,&nbsp;Hyoung Jung Kim ,&nbsp;Jin Hee Kim ,&nbsp;Ki Byung Song ,&nbsp;Jae Hoon Lee ,&nbsp;Dae Wook Hwang","doi":"10.1016/j.ejrad.2026.112735","DOIUrl":"10.1016/j.ejrad.2026.112735","url":null,"abstract":"<div><h3>Objective</h3><div>To evaluate magnetic resonance imaging (MRI) characteristics that differentiate large-duct pancreatic ductal adenocarcinoma (LD-PDAC) from malignant intraductal papillary mucinous neoplasm (IPMN).</div></div><div><h3>Materials and Methods</h3><div>We retrospectively analyzed preoperative MRI data from 42 LD-PDAC patients, 201 malignant IPMN patients (166 with high-grade dysplasia and 35 with invasive carcinoma), and 8 LD-PDAC arising from IPMN patients. Two radiologists independently assessed MRI features including tumor morphology and accompanying imaging features. Multivariable logistic regression, diagnostic performance of combined imaging predictors, and survival outcomes across disease entities were evaluated.</div></div><div><h3>Results</h3><div>LD-PDAC predominantly appeared as solid-dominant tumors, either solid masses with internal cystic portions (69.0%) or pure solid masses (16.7%). Malignant IPMN presented mainly as cyst-dominant tumors, either pure cystic masses (55.2%) or cystic masses with internal solid components (42.8%). Multivariable logistic regression analysis identified solid-dominant tumor morphology (odds ratio [OR], 77.89; 95% confidence interval [CI], 4.94–1229.16), peripancreatic infiltration (OR, 34.47; 95% CI, 2.49–476.79), and absence of disproportionate pancreatic duct dilatation (OR, 0.06; 95% CI, 0.01–0.59) as independent imaging features favoring LD-PDAC. LD-PDAC showed significantly shorter overall and recurrence-free survival than malignant IPMN (p &lt; 0.001), while survival did not differ significantly between LD-PDAC and IPMN with associated invasive carcinoma. Overall, 26.2% of LD-PDAC cases were misdiagnosed, mainly due to misinterpretation of cyst wall characteristics and atypical duct dilatation.</div></div><div><h3>Conclusion</h3><div>MRI may aid in differentiating LD-PDAC from malignant IPMN by integrating tumor morphology and accompanying imaging features, including pancreatic ductal dilatation patterns and peripancreatic infiltration; however, substantial imaging overlap persists, resulting in a clinically meaningful misdiagnosis rate.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112735"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147270181","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Gallbladder adenomyomatosis revisited – Does size matter? is follow-up required for large lesions?","authors":"Shirley Shechter,&nbsp;Dana Ben-Ami Shor,&nbsp;Roie Tzadok,&nbsp;Hila Yashar,&nbsp;Sapir Lazar,&nbsp;Yuval Katz,&nbsp;Arthur Chernomorets,&nbsp;Rivka Kessner","doi":"10.1016/j.ejrad.2026.112698","DOIUrl":"10.1016/j.ejrad.2026.112698","url":null,"abstract":"<div><h3>Objectives</h3><div>Adenomyomatosis (ADM) is generally considered a benign condition. However, it can be associated with chronic cholecystitis − a known risk factor for gallbladder cancer. Therefore, studies have proposed follow-up with ultrasound for asymptomatic patients with focal ADM. Currently, there are no formal recommendations regarding the frequency and length of follow-up. The aims of this study were to assess the growth of ADM lesions during follow-up and to examine the differences between larger and smaller ADM lesions.</div></div><div><h3>Methods</h3><div>144 patients who underwent MRI-MRCP at our institution between the years 2014–2024 were identified through radiological reports as having a diagnosis of ADM. 43 patients had more than one examination. Demographic, clinical and radiological data were collected retrospectively. We divided the cohort into two groups based on the primary lesion size (axial diameter below or above 1.5 cm) and compared between them.</div></div><div><h3>Results</h3><div>The group of small lesions included 98 patients and the larger lesions group included 46 patients. We did not find a statistically significant correlation between the size of.</div><div>ADM and the demographic or clinical characteristics examined. Only 9 ADM lesions grew during follow-up − 6 from the smaller lesions group and 3 from the large lesions group (p &gt; 0.05). The median follow-up period was 35 months. None of our patients developed gallbladder carcinoma.</div></div><div><h3>Conclusions</h3><div>Our results confirm the common hypothesis that ADM are benign lesions. Therefore, we believe that follow-up is not needed for lesions with a clear diagnosis of focal ADM.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112698"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146200556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Interobserver variability of recall decisions between mammography readers in the English NHS breast screening programme: A comparison of interobserver variability measures","authors":"Laura Quinn ,&nbsp;David Jenkinson ,&nbsp;Sian Taylor-Phillips ,&nbsp;Yemisi Takwoingi ,&nbsp;Alice Sitch","doi":"10.1016/j.ejrad.2026.112723","DOIUrl":"10.1016/j.ejrad.2026.112723","url":null,"abstract":"<div><h3>Objectives</h3><div>To evaluate interobserver variability between mammogram readers’ recall decisions in the English NHS breast screening programme, comparing different variability measures.</div></div><div><h3>Methods</h3><div>Data from 401,682 women in 22 NHS centres who underwent mammographic screening interpreted independently by two mammogram readers were included. Percentage agreement, prevalence-adjusted bias-adjusted-kappa (PABAK), Gwet’s agreement coefficient (Gwet’s AC) and Cohen’s kappa were reported with 95% confidence intervals. Analyses were performed separately for women at first and subsequent screening appointments, by cancer diagnosis, reader recall rates and age group.</div></div><div><h3>Results</h3><div>Of 86,287 women at first screening, 6,491 (7.5%) were recalled, compared to 9,488 (3.0%) of 315,395 at subsequent screenings. Percentage agreement, Gwet’s AC, and PABAK were lower for first screening than subsequent (93.6%, 95%CI: 93.4–93.7 vs 97.2%, 95%CI: 97.2–97.3), (92.3, 95%CI:92.1 to 92.5 vs 97.0, 95% CI: 97.0 to 97.1) and (87.2, 95%CI: 86.9–87.4 vs 94.4, 95%CI: 94.3–94.5), whereas Cohen’s kappa, which is biased downwards when prevalence of recall is lower, did not change (61.6, 95%CI: 60.7–62.5 vs 61.8, 95%CI: 61.0–62.5). Percentage agreement, Gwet’s AC, and PABAK were lower for women with cancer detected than without, but Cohen’s kappa showed the opposite pattern, driven by prevalence bias. Percentage agreement, Gwet’s AC, and PABAK were lower when one/both readers had high recall rates, but Cohen’s kappa showed no important pattern.</div></div><div><h3>Conclusions</h3><div>Percentage agreement, Gwet’s AC, and PABAK showed lower agreement for interpreting the more challenging first screen, without assistance of previous mammograms, when women had cancer and when one/both readers had a high recall rate. Cohen’s kappa was heavily distorted by outcome prevalence. Despite widespread use, Cohen’s kappa is inappropriate for low prevalence settings such as screening, or making comparisons when prevalence varies.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112723"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146161987","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The application of Machine learning in predicting the outcomes of minimally invasive treatments for uterine Fibroids: A systematic review and meta-analysis","authors":"Mohammad-Reza Hosseini-Siyanaki ,&nbsp;Seyyed Mohammad Hosseini ,&nbsp;Maryam Afshari ,&nbsp;Fatemeh Kanaani Nejad ,&nbsp;Hoda Mehrabi ,&nbsp;Reza Elahi ,&nbsp;Ahmadreza Sohrabi-Ashlaghi ,&nbsp;Babak Ahmadi ,&nbsp;Shakiba Houshi ,&nbsp;Fereshteh Yazdanpanah ,&nbsp;Zahra Beyzavi ,&nbsp;Shams Iqbal","doi":"10.1016/j.ejrad.2026.112726","DOIUrl":"10.1016/j.ejrad.2026.112726","url":null,"abstract":"<div><h3>Rationale and Objectives</h3><div>Uterine fibroids (UFs) are common benign tumors that impact women’s health, particularly through symptoms such as abnormal bleeding or reproductive dysfunction. Interventional radiology (IR) techniques like uterine artery embolization (UAE) and high-intensity focused ultrasound (HIFU) are minimally invasive alternatives to surgery. Machine learning (ML) has shown promise in predicting treatment outcomes, though the optimal model remains uncertain. This systematic review and <em>meta</em>-analysis evaluate models predicting outcomes of minimally invasive treatments for uterine fibroids.</div></div><div><h3>Materials &amp; Methods</h3><div>A comprehensive search was conducted across five databases (PubMed, Embase, Scopus, Web of Science, and Cochrane) through November 2024, following PRISMA guidelines and registered in PROSPERO. Studies using ML to predict different outcomes of UFs treatment via minimally invasive treatments were included. PROBAST + AI was used to assess study quality. Pooled sensitivity, specificity, and AUC values were calculated using a bivariate random effect model.</div></div><div><h3>Results</h3><div>Out of 1,114 records, fourteen studies met the inclusion criteria, with 12 focusing on HIFU and two on UAE. Logistic regression was the most commonly used approach, while gradient‑boosting models reported high discrimination in some individual studies; however, external validation was uncommon and risk of bias was frequently high. AUCs for radiomics-based models ranged from 0.668 to 0.887, and combined models ranged from 0.773 to 0.93. Meta-analysis of five HIFU-based radiomics studies demonstrate pooled sensitivity of 75% and specificity of 76% respectively, with an AUC of 0.82.</div></div><div><h3>Conclusion</h3><div>ML models, particularly those integrating radiomics and clinical data, show strong performance in predicting image-guided treatment outcomes in UFs. These approaches support a promising path toward individualized treatment planning and may improve patient selection in clinical workflow.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112726"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146200578","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Explainable and evidence-linked recommendations for spine surgery via a retrieval-augmented LLM agent","authors":"Yiren Li,&nbsp;Mingyu Lv,&nbsp;Liang Cheng,&nbsp;Yongxu Xie,&nbsp;Qian Gu,&nbsp;Haozheng He,&nbsp;Zhuguang Chen,&nbsp;Duopei Fang,&nbsp;Xiang Zhou","doi":"10.1016/j.ejrad.2026.112734","DOIUrl":"10.1016/j.ejrad.2026.112734","url":null,"abstract":"<div><h3>Background</h3><div>Clinical decision-making in spinal and spinal cord diseases requires a comprehensive assessment of imaging findings, neurological status, bone integrity, and patient-centered goals. Recently, the emergence of Large Language Models (LLMs) has provided new tools for intelligent decision support; however, their reliability and clinical interpretability remain to be systematically evaluated.</div></div><div><h3>Methods</h3><div>We propose a novel retrieval augmented generation (RAG)-based framework specifically tailored for spinal disease decision support and systematically evaluated four advanced LLM agents (Gemini 2.5 Flash, DeepSeek, GPT-4o, GPT-4o-mini). The framework integrates domain-specific prompting, structured response formatting, and evidence citation tracking.We used 200 real-world spinal cases, each involving diagnostic, therapeutic, and follow-up tasks. Five spine surgeons independently evaluated model outputs using an 11-dimension rubric; each dimension rated on a 5-point Likert scale to evaluate both clinical and technical performance. Inter-group differences were analyzed using the Kruskal–Wallis and Dunn’s tests (P &lt; 0.05), with radar plots used for multidimensional visualization.</div></div><div><h3>Results</h3><div>The proposed expert-evaluated framework enables a comprehensive, real-case-based comparison of four LLM agents. Gemini 2.5 Flash achieved the highest overall score (49.25 ± 2.88), significantly outperforming DeepSeek (47.49 ± 3.34, <em>P</em> &lt; 0.001), GPT-4o (45.59 ± 3.89, <em>P</em> &lt; 0.001), and GPT-4o-mini (41.23 ± 5.96, <em>P</em> &lt; 0.001). It demonstrated leading performance particularly in humanistic care, follow-up suggestion, and test recommendation. DeepSeek showed superior capability in differential completeness (mean = 4.76), significantly outperforming the other three models (<em>P</em> &lt; 0.001). Although GPT-4o-mini demonstrated stable system performance (mean = 4.61), it underperformed in core clinical reasoning dimensions. These findings reveal substantial inter-model variability in spine-specific clinical reasoning, an aspect often overlooked in prior non-benchmark LLM evaluations.</div></div><div><h3>Conclusion</h3><div>Among the four evaluated LLM agents, Gemini 2.5 Flash and DeepSeek demonstrated superior clinical accuracy, comprehensiveness, and usability in spine-related decision support. These findings support the potential of domain-adapted RAG agents to enhance evidence-based spinal care by providing accurate and comprehensive decision support. Future research should focus on multimodal integration data (e.g., imaging and clinical notes) and conducting prospective validation in real-world clinical environments.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112734"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146776188","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and validation of a nomogram to predict surgical resection after conversion therapy in unresectable hepatocellular carcinoma","authors":"Yuhao Su ,&nbsp;Yuxin Liang ,&nbsp;Deyuan Zhong,&nbsp;Yahui Chen,&nbsp;Hongtao Yan,&nbsp;Qinyan Yang,&nbsp;Ming Wang,&nbsp;Zhengwei Leng,&nbsp;Xiaolun Huang","doi":"10.1016/j.ejrad.2026.112724","DOIUrl":"10.1016/j.ejrad.2026.112724","url":null,"abstract":"<div><h3>Objective</h3><div>This study aimed to explore factors associated with the likelihood of surgical resection after triple-combination conversion therapy in patients with initially unresectable hepatocellular carcinoma (uHCC) and to develop an exploratory predictive model.</div></div><div><h3>Methods</h3><div>A retrospective analysis was conducted using clinical data from 210 patients with uHCC who underwent triple-combination conversion therapy at Sichuan Cancer Hospital between January 2022 and January 2025. Patients were randomly assigned to a training cohort (n = 147) and a validation cohort (n = 63) in a 7:3 ratio. Least absolute shrinkage and selection operator (LASSO) regression was applied to screen candidate predictors, followed by multivariate logistic regression to identify factors associated with surgical conversion. A nomogram was constructed based on these variables, and its discriminative ability, calibration, and potential clinical utility were internally assessed using receiver operating characteristic (ROC) analysis, calibration plots, the Hosmer–Lemeshow test, and decision curve analysis (DCA).</div></div><div><h3>Results</h3><div>Among the 210 patients, 47 (22.4%) successfully underwent conversion and radical resection. Multivariate logistic regression analysis suggested that lower tumor burden score (TBS; OR = 0.663), lower neutrophil-to-lymphocyte ratio (NLR; OR = 0.572), lower C-reactive protein-to-albumin ratio (CAR; OR = 0.057), and absence of cirrhosis (OR = 0.289) were associated with a higher likelihood of successful surgical conversion (P &lt; 0.05). The nomogram showed moderate to good discriminative performance, with areas under the ROC curve (AUCs) of 0.850 (95% CI: 0.784–0.915) in the training cohort and 0.871 (95% CI: 0.783–0.962) in the validation cohort. Calibration plots and decision curve analysis provided descriptive information regarding model performance within the study cohort.</div></div><div><h3>Conclusion</h3><div>The proposed nomogram, incorporating TBS, NLR, CAR, and cirrhosis status, represents an exploratory tool for estimating the probability of surgical conversion following triple-combination therapy in patients with uHCC. While the model may provide supplementary information to support clinical assessment and patient stratification, further multicenter and prospective studies are required to externally validate and refine its performance before broader clinical application.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"197 ","pages":"Article 112724"},"PeriodicalIF":3.3,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146219026","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}