QVQuentin Vanderbecq

Quentin Vanderbecq

*This is an algorithmically generated author record based on publicly available information.

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


Quentin Vanderbecq has a total of 53 co-authors. Here are the top co-authors by number of shared publications:

  • Marc Zins

    Department of Medical Imaging, Saint Joseph Hospital, Paris, France

    3 shared publications
  • Mathilde Wagner

    Department of Radiology, Assistance Publique–Hôpitaux de Paris (AP-HP), Sorbonne Université, Saint-Antoine Hospital, 184 rue du Faubourg Saint-Antoine, 75012 Paris, France

    3 shared publications
  • Mathilde Wagner

    Department of Radiology, AP-HP Sorbonne Université, Paris, France

    3 shared publications
  • Emilie Chouzenoux

    CVN, CentraleSupélec, INRIA Paris Saclay, Université Paris-Saclay, Gif-Sur-Yvette, France

    2 shared publications
  • Jean-Christophe Pesquet

    CVN, CentraleSupélec, INRIA Paris Saclay, Université Paris-Saclay, Gif-Sur-Yvette, France

    2 shared publications
  • Lionel Arrive

    Department of Radiology, AP-HP.Sorbonne, Saint Antoine Hospital, Paris, France

    2 shared publications

Latest Publications

  • Open Access

    Use of Large Language Models for Automated Short-Answer Question Grading in Second-Cycle Medical Education: Performance and Consistency Evaluation Study

    by Sabrina Sidali, Pierre-Antoine Soret, Mathilde Wagner, Thibault Voron, Quentin Vanderbecq

    Background: Short-answer questions (SAQs) effectively assess higher-order reasoning skills in medical education but face challenges in scalability due to time-intensive manual grading and concerns about interrater reliability. Recent advances in large language models (LLMs) may offer a solution through automated assessment. Objective: This study evaluated the performance and consistency of LLMs—ChatGPT, Claude Sonnet 4, and Mistral Pro—compared with those of human graders for SAQ grading in hepatology and gastroenterology. Methods: We retrospectively analyzed SAQs from hepatology and gastroenterology examinations at Sorbonne University, Paris, France (2023-2025). Three human evaluators (2 hepatogastroenterologists and 1 radiologist) independently graded all responses twice, with the majority consensus score serving as the reference standard. Six LLMs were evaluated: ChatGPT (GPT-4o, GPT-4.5, GPT-4o3, and GPT-5), Claude Sonnet 4, and Mistral Pro. Each LLM graded all responses twice using standardized prompts, including questions, model answers, grading guidelines, and scoring rubrics. Agreement was assessed using quadratic weighted kappa (QWK), Cohen κ, the intraclass correlation coefficient (ICC), and the mean absolute error (MAE) for interhuman, intrahuman, LLM-versus-human, and intramodel comparisons. Results: A total of 902 student responses from 12 SAQs were included and analyzed. Interhuman grader agreement was almost perfect (QWK 0.90-0.92; ICC 0.90-0.92; MAE 0.05-0.07). Claude Sonnet 4 and GPT-4o3 demonstrated near-human–level performance when compared with the majority vote of human evaluators (Claude Sonnet 4: QWK 0.88-0.90; ICC 0.88-0.90; MAE 0.06-0.07 and GPT-4o3: QWK 0.82-0.89; ICC 0.82-0.89; MAE 0.07-0.09). GPT-5 showed substantial agreement (QWK 0.77-0.79). Mistral Pro and older ChatGPT versions performed poorly (QWK <0.63). Intramodel consistency was the highest for Claude Sonnet 4 (QWK 0.93), approaching the best human intrarater reliability. Agreement did not correlate with prompt design choices (number of examples or minimum word count). Conclusions: Advanced LLMs, particularly Claude Sonnet 4 and GPT-4o3, demonstrated near-human–level grading performance, supporting their potential use as assistive tools in SAQ grading. However, performance varied by model, and agreement alone does not ensure educational validity. Until broader multicenter validation is conducted across diverse domains, stakes, and question formats—along with safeguards for bias, privacy, and model drift—LLMs should augment rather than replace human faculty graders.


  • Open Access

    Early imaging signs associated with subsequent diagnosis of primary sclerosing cholangitis

    by Quentin Vanderbecq, Pauline Iorio, Pierre-Antoine Soret, Sanaa El Mouhadi, Eleonora De Martin, Lionel Arrivé, Sara Lemoinne, Maïté Lewin, Mathilde Wagner

    BACKGROUND: Primary sclerosing cholangitis (PSC) is characterized by multifocal biliary strictures on MRI/MRCP. However, we have observed that young PSC patients often displayed a marked intrahepatic bile duct visibility, sometimes with gallbladder enlargement, before developing biliary strictures. This study aimed to evaluate whether this imaging pattern was associated with subsequent diagnosis of PSC. PATIENTSAND METHODS: This retrospective two-center study included patients younger than 21 years who had a baseline liver MRI/MRCP, without biliary strictures. The endpoint was a definite diagnosis of large duct PSC during follow-up. The association of baseline imaging findings with the endpoint was evaluated by multivariable logistic regression, Cox models, and Kaplan-Meier analyses. RESULTS: 94 patients were included: 54% women, mean age of 16, 36% with IBD. Over a mean follow-up of 7.75 years, 21 patients were subsequently diagnosed with large duct PSC after a median time of 5.60 years. Marked intrahepatic bile duct visibility and gallbladder enlargement displayed good performance to predict the subsequent diagnosis of PSC (AUC, 0.77 and 0.69, respectively). The presence of both signs displayed a specificity of 1.00. In multivariable analyses, marked intrahepatic bile duct visibility and gallbladder enlargement were independently associated with subsequent PSC diagnosis (OR, 13.35; 95% CI, 3.00– 59.41; p < 0.001 and OR, 12.36; 95% CI, 1.93–79.09; p = 0.008, respectively) and shorter time to diagnosis (HR, 6.42; 95% CI, 1.99–20.75; p = 0.002 and HR, 3.36; 95% CI, 1.22–9.27; p = 0.019, respectively). Survival analyses confirmed the association between these two imaging signs and subsequent PSC diagnosis (HR, 7.78; 95% CI, 2.62–23.16; p < 0.001 and HR, 3.85; 95% CI, 1.64–9.08; p < 0.001, respectively). CONCLUSIONS: In young patients without biliary strictures, marked intrahepatic bile duct visibility and gallbladder enlargement on MRI/MRCP were strongly associated with subsequent diagnosis of large duct PSC. IMPACT AND IMPLICATIONS: Diagnosis of PSC relies on the presence of multifocal biliary strictures. However, early diagnosis of PSC is crucial since it could trigger earlier and more effective therapeutic interventions. In young patients without biliary strictures, we identified two imaging signs significantly associated with the subsequent diagnosis of large duct PSC: marked intrahepatic bile duct visibility and gallbladder enlargement. These findings should encourage closer monitoring of young patients who display these imaging patterns. HIGHLIGHTS: Marked intrahepatic bile duct visibility can predict the subsequent diagnosis of large duct PSC. Gallbladder enlargement can predict the subsequent diagnosis of large duct PSC. These two signs were significantly associated with subsequent diagnosis of large duct PSC.

  • Radiologist-AI Collaboration for Ischemia Diagnosis in Small-Bowel Obstruction: Multicentric Development and External Validation of a Multimodal Deep Learning Model

    by Quentin Vanderbecq, Wen Fan Xia, Emilie Chouzenoux, Dima Smaily, Jean-Christophe Pesquet, Marc Zins, Mathilde Wagner

    This study aims to develop and externally validate a multimodal AI model for detecting ischemia complicating small-bowel obstruction (SBO). We combined 3D CT data with routine laboratory markers (C-reactive protein, neutrophil count) and, optionally, radiology report indication/history text. From two centers, 1350 CT examinations were curated; 771 confirmed SBO scans were used for model development with patient-level splits. Ischemia labels were defined by surgical confirmation within 24 h of imaging. Models (MViT, ResNet-101, DaViT) were trained as unimodal and multimodal variants. External testing was used for 66 independent cases from a third center. Four radiologists (two residents and two experts) read the test set with and without AI assistance. Performance was assessed using AUC, sensitivity, specificity, and 95% bootstrap confidence intervals; predictions included a confidence score. The image-plus-laboratory model performed best on external testing (AUC 0.69 [0.59–0.79], sensitivity 0.89 [0.76–1.00], and specificity 0.44 [0.35–0.54]). Adding report text improved internal validation but did not generalize externally; image + text and full multimodal variants did not exceed image + laboratory performance. Across readers, baseline AUC ranged from 0.496 [0.361–0.640] to 0.745 [0.589–0.875] and increased with reader experience. With AI assistance, AUC ranged from 0.565 [0.419–0.717] to 0.845 [0.714–0.952] and from 0.519 [0.373–0.669] to 0.845 [0.708–0.954] when confidence scores were displayed, showing consistent but non-significant changes whatever the experience level. A multimodal model combining CT and lab data surpassed unimodal approaches for 24-h ischemia detection; as a triage-support tool, it showed a consistent but non-significant improvement in radiologist performance.


  • Développement et validation de modèles basés sur la tomodensitométrie pour l’aide au diagnostic et à la prise en charge de l’occlusion intestinale

    by Quentin Vanderbecq

    L'occlusion aiguë est une urgence fréquente et hétérogène. Nous avons entraîné et testé, sur plusieurs centaines de tomodensitométries, différents modèles 3D pour : (1) détecter l'occlusion (oui/non), (2) distinguer occlusion mécanique vs fonctionnelle, (3) localiser les occlusions sur bride par segmentation de la zone de transition, (4) prédire la chirurgie (urgente ≤24 h / différée / aucune), et (5) détecter l'ischémie de la paroi de l'intestin grêle confirmée au bloc. Le réseau neuronal convolutif mixte 2D-3D atteint une précision équilibrée jusqu'à 0,89 pour la détection binaire, avec une sensibilité proche de 0,98 en validation externe. Si les performances restent limitées pour la différenciation occlusions mécaniques/fonctionnelles ou la localisation des brides, plusieurs approches montrent un potentiel prometteur, notamment les modèles multimodaux. L'intégration des données biologiques (CRP, neutrophiles) dans un modèle Vision Transformer améliore nettement la détection de l'ischémie précoce en cas d'occlusion du grêle (Se : 0,76 →0,89 ; AUC : 0.53→0,69). Les modèles de fondation (MAISI, CT-FM) testés en few-shot learning n'ont pas encore dépassé les modèles classiques, mais posent des bases intéressantes pour de futures recherches. Par ailleurs, certains modèles se révèlent utiles pour améliorer le triage et accompagner les radiologues dans l'évaluation de la souffrance digestive. Enfin, ces travaux montrent l'intérêt d'augmenter et diversifier les données, d'affiner les annotations (ischémie, bride, transition) et d'explorer des architectures plus hiérarchiques et plus spécifique à l'imagerie 3D. L'IA, bien que perfectible, s'affirme déjà comme un soutien précieux dans l'aide au diagnostic des occlusions intestinales.

  • Open Access

    Evaluating MRI response criteria in microsatellite instability-high rectal cancer treated with immune checkpoint inhibitors

    by Q. Vanderbecq, R. Cohen, M. Camus, X. Dray, Y. Parc, T. André, M. Wagner


  • Open Access

    Use of Large Language Models for Automated Short-Answer Question Grading in Second-Cycle Medical Education: Performance and Consistency Evaluation Study (Preprint)

    by Sabrina Sidali, Pierre-Antoine Soret, Mathilde Wagner, Thibault Voron, Quentin Vanderbecq

    BACKGROUND Short-answer questions (SAQs) effectively assess higher-order reasoning skills in medical education but face challenges in scalability due to time-intensive manual grading and concerns about interrater reliability. Recent advances in large language models (LLMs) may offer a solution through automated assessment. OBJECTIVE This study evaluated the performance and consistency of LLMs—ChatGPT, Claude Sonnet 4, and Mistral Pro—compared with those of human graders for SAQ grading in hepatology and gastroenterology. METHODS We retrospectively analyzed SAQs from hepatology and gastroenterology examinations at Sorbonne University, Paris, France (2023-2025). Three human evaluators (2 hepatogastroenterologists and 1 radiologist) independently graded all responses twice, with the majority consensus score serving as the reference standard. Six LLMs were evaluated: ChatGPT (GPT-4o, GPT-4.5, GPT-4o3, and GPT-5), Claude Sonnet 4, and Mistral Pro. Each LLM graded all responses twice using standardized prompts, including questions, model answers, grading guidelines, and scoring rubrics. Agreement was assessed using quadratic weighted kappa (QWK), Cohen κ, the intraclass correlation coefficient (ICC), and the mean absolute error (MAE) for interhuman, intrahuman, LLM-versus-human, and intramodel comparisons. RESULTS A total of 902 student responses from 12 SAQs were included and analyzed. Interhuman grader agreement was almost perfect (QWK 0.90-0.92; ICC 0.90-0.92; MAE 0.05-0.07). Claude Sonnet 4 and GPT-4o3 demonstrated near-human–level performance when compared with the majority vote of human evaluators (Claude Sonnet 4: QWK 0.88-0.90; ICC 0.88-0.90; MAE 0.06-0.07 and GPT-4o3: QWK 0.82-0.89; ICC 0.82-0.89; MAE 0.07-0.09). GPT-5 showed substantial agreement (QWK 0.77-0.79). Mistral Pro and older ChatGPT versions performed poorly (QWK <0.63). Intramodel consistency was the highest for Claude Sonnet 4 (QWK 0.93), approaching the best human intrarater reliability. Agreement did not correlate with prompt design choices (number of examples or minimum word count). CONCLUSIONS Advanced LLMs, particularly Claude Sonnet 4 and GPT-4o3, demonstrated near-human–level grading performance, supporting their potential use as assistive tools in SAQ grading. However, performance varied by model, and agreement alone does not ensure educational validity. Until broader multicenter validation is conducted across diverse domains, stakes, and question formats—along with safeguards for bias, privacy, and model drift—LLMs should augment rather than replace human faculty graders.

  • Open Access

    Radiologist-AI Collaboration for Ischemia Diagnosis in Small Bowel Obstruction: Multicentric Development and External Validation of a Multimodal Deep Learning Model

    by Quentin Vanderbecq, Wen Fan Xia, Émilie Chouzenoux, Jean-Christophe Pesquet, Marc Zins, Mathilde Wagner

    Purpose To develop and externally validate a multimodal AI model for detecting ischaemia complicating small-bowel obstruction (SBO). Methods We combined 3D CT data with routine laboratory markers (C-reactive protein, neutrophil count) and, optionally, radiology report text. From two centers, 1,350 CT examinations were curated; 771 confirmed SBO scans were used for model development with patient-level splits. Ischemia labels were defined by surgical confirmation within 24 hours of imaging. Models (MViT, ResNet-101, DaViT) were trained as unimodal and multimodal variants. External testing was used for 66 independent cases from a third center. Two radiologists (attending, resident) read the test set with and without AI assistance. Performance was assessed using AUC, sensitivity, specificity, and 95% bootstrap confidence intervals; predictions included a confidence score. Results The image-plus-laboratory model performed best on external testing (AUC 0.69 [0.59–0.79], sensitivity 0.89 [0.76–1.00], and specificity 0.44 [0.35–0.54]). Adding report text improved internal validation but did not generalize externally; image+text and full multimodal variants did not exceed image+laboratory performance. Without AI, the attending outperformed the resident (AUC 0.745 [0.617–0.845] vs 0.706 [0.581–0.818]); with AI, both improved, attending 0.752 [0.637–0.853] and resident 0.752 [0.629–0.867], rising to 0.750 [0.631–0.839] and 0.773 [0.657–0.867] with confidence display; differences were not statistically significant. Conclusion A multimodal AI that combines CT images with routine laboratory markers outperforms single-modality approaches and boosts radiologist readers’ performance notably junior, supporting earlier, more consistent decisions within the first 24 hours. Key Points A multimodal artificial intelligence (AI) model that combines CT images with laboratory markers detected ischemia in small-bowel obstruction with AUC 0.69 (95% CI 0.59–0.79) and sensitivity 0.89 (0.76–1.00) on external testing, outperforming single-modality models. Adding report text did not generalize across sites: the image+text model fell from AUC 0.82 (internal) to 0.53 (external), and adding text to image+biology left external AUC unchanged (0.69) with similar specificity (0.43–0.44). With AI assistance both junior and senior readers improved; the junior’s AUC rose from 0.71 to 0.77, reaching senior-level performance. Summary Statement A multicentric AI model combining CT and routine laboratory data (CRP and neutrophilia) improved radiologists’ detection of ischemia in small-bowel obstruction. This tool supports earlier decision-making within the first 24 hours.


  • Deep learning for automatic bowel-obstruction identification on abdominal CT

    by Quentin Vanderbecq, Maxence Gelard, Jean-Christophe Pesquet, Mathilde Wagner, Lionel Arrive, Marc Zins, Emilie Chouzenoux

    Rationale and objectives Automated evaluation of abdominal computed tomography (CT) scans should help radiologists manage their massive workloads, thereby leading to earlier diagnoses and better patient outcomes. Our objective was to develop a machine-learning model capable of reliably identifying suspected bowel obstruction (BO) on abdominal CT. Materials and methods The internal dataset comprised 1345 abdominal CTs obtained in 2015–2022 from 1273 patients with suspected BO; among them, 670 were annotated as BO yes/no by an experienced abdominal radiologist. The external dataset consisted of 88 radiologist-annotated CTs. We developed a full preprocessing pipeline for abdominal CT comprising a model to locate the abdominal-pelvic region and another model to crop the 3D scan around the body. We built, trained, and tested several neural-network architectures for the binary classification (BO, yes/no) of each CT. F1 and balanced accuracy scores were computed to assess model performance. Results The mixed convolutional network pretrained on a Kinetics 400 dataset achieved the best results: with the internal dataset, the F1 score was 0.92, balanced accuracy 0.86, and sensitivity 0.93; with the external dataset, the corresponding values were 0.89, 0.89, and 0.89. When calibrated on sensitivity, this model produced 1.00 sensitivity, 0.84 specificity, and an F1 score of 0.88 with the internal dataset; corresponding values were 0.98, 0.76, and 0.87 with the external dataset. Conclusion The 3D mixed convolutional neural network developed here shows great potential for the automated binary classification (BO yes/no) of abdominal CT scans from patients with suspected BO. Clinical relevance statement The 3D mixed CNN automates bowel obstruction classification, potentially automating patient selection and CT prioritization, leading to an enhanced radiologist workflow. Key Points • Bowel obstruction’s rising incidence strains radiologists. AI can aid urgent CT readings. • Employed 1345 CT scans, neural networks for bowel obstruction detection, achieving high accuracy and sensitivity on external testing. • 3D mixed CNN automates CT reading prioritization effectively and speeds up bowel obstruction diagnosis.


  • Radiologie diagnostique et thérapeutique de la pancréatite aiguë grave

    by Quentin Vanderbecq, Sanaa El Mouhadi, Amal Ali, Lise Minssen, Lionel Arrive

  • Improving pain control during transarterial chemoembolization for hepatocellular carcinoma performed under local anesthesia with multimodal analgesia

    by Quentin Vanderbecq, Jules Grégory, Jeremy Dana, Marco Dioguardi Burgio, Lorenzo Garzelli, Lucas Raynaud, Sébastien Frémy, Luisa Paulatto, Mohamed Bouattour, Juliette Kavafyan-Lasserre, et al.