{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,28]],"date-time":"2026-07-28T11:58:57Z","timestamp":1785239937834,"version":"3.55.0"},"reference-count":55,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,7,28]],"date-time":"2026-07-28T00:00:00Z","timestamp":1785196800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Radiol."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Automated chest x-ray reporting could substantially reduce the burden on radiology services worldwide; however, the implementation of current vision\u2013language models (VLMs) in clinical workflows remains limited by factual errors, hallucinations, and inadequate clinical reliability. Bridging this implementation gap requires frameworks that are not only technically sound but also designed for safe integration into real-world healthcare settings.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Objective<\/jats:title>\n                    <jats:p>This study aimed to improve the clinical accuracy and factual consistency of radiology report generation by introducing an agent-inspired, three-stage reasoning framework and evaluating its feasibility as research prototype for potential implementation in resource-constrained and high-throughput clinical environments.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We propose the Swin-Qwen3 vision\u2013language architecture, which integrates a Swin Transformer visual encoder with a Q-Former-style query-driven cross-attention module aligned with a large language model (Qwen3-0.6B). A three-stage generation strategy\u2014comprising initial report drafting, clinical verification, and structured refinement\u2014was guided by role-specific prompts to simulate distinct clinical reasoning behaviours. Parameter-efficient fine-tuning via low-rank adaptation (LoRA) enabled training within standard GPU constraints. The framework was evaluated on the full IU x-Ray test set (321 Samples) using lexical, semantic, clinical, and factuality metrics and compared with one- and two-stage ablations. Computational feasibility and inference overhead were assessed for offline or batch processing contexts.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      The three-stage framework showed modest but consistent improvements over ablation baselines. CheXpert-F1 reached 0.7038 (compared to 0.7154 for one-stage and 0.7123 for two-stage), Clinical-F1 reached 0.5401 (compared to 0.5478 for one-stage and 0.5449 for two-stage), and RadGraph-F1 scored 0.5467 (compared to 0.5478 for one-stage and 0.5445 for two-stage). The factuality score reached 0.843 (compared to 0.845 for one-stage and 0.845 for two-stage), while the hallucination rate remained at 0.6116 (compared to 0.6109 for one-stage and 0.6109 for two-stage). Clinical recall decreased from 0.8361 (one-stage) and 0.8337 (two-stage) to 0.8044 (three-stage), reflecting a trade-off between sensitivity and precision. However, the clinical verifier demonstrated substantial effectiveness, correcting 85.6% of identified hallucinations. Lexical quality improved: BLEU-4 was 0.0628 (+10.9% vs. one-stage; +13.1% vs. two-stage), ROUGE-2-F was 0.0957 (+10.2% vs. one-stage; +13.1% vs. two-stage), and METEOR was 0.3728 (+4.5% vs. one-stage; +6.3% vs. two-stage). Improvements in METEOR (\n                      <jats:italic>p<\/jats:italic>\n                      \u2009&amp;lt;\u20090.001) and ROUGE-2-F (\n                      <jats:italic>p<\/jats:italic>\n                      \u2009&amp;lt;\u20090.05) were statistically significant. A 14.8\u2009\u00d7\u2009 inference overhead was identified, indicating suitability for offline reporting contexts.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>The Swin-Qwen3 three-stage framework demonstrates that explicit clinical verification and iterative refinement can modestly enhance the clinical reliability of VLMs for automated radiology reporting. The parameter-efficient design is scalable across diverse clinical settings, including resource-limited environments, and supports a practical deployment pathway aligned with SDGs 3 (Good Health and Well-being), 9 (Industry, Innovation and Infrastructure), and 17 (Partnerships for the Goals). However, the residual hallucination rate of 61.2% indicates that the current system is a promising research prototype rather than a clinically deployable tool. Future work should prioritize multi-institutional validation on larger datasets (e.g., MIMIC-CXR) with radiologist expert review, regulatory evaluation, and prospective clinical integration studies.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fradi.2026.1875905","type":"journal-article","created":{"date-parts":[[2026,7,28]],"date-time":"2026-07-28T11:10:17Z","timestamp":1785237017000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Swin-Qwen3: a three-stage vision\u2013language framework for automated radiology report generation with multi-agent verification"],"prefix":"10.3389","volume":"6","author":[{"given":"Hamida","family":"Abdaoui","sequence":"first","affiliation":[{"name":"Laboratory of Biophysics and Medical Technologies, Higher Institute of Medical Technologies of Tunis (ISTMT), University of Tunis El Manar","place":["Tunis, Tunisia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sabri","family":"Barbaria","sequence":"additional","affiliation":[{"name":"Laboratory of Biophysics and Medical Technologies, Higher Institute of Medical Technologies of Tunis (ISTMT), University of Tunis El Manar","place":["Tunis, Tunisia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ahmed","family":"Al Kuwaiti","sequence":"additional","affiliation":[{"name":"Department of Dental Education, College of Dentistry, Deanship of Quality and Academic Accreditation, Imam Abdulrahman Bin Faisal University","place":["Dammam, Saudi Arabia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Noureddine","family":"Raouafi","sequence":"additional","affiliation":[{"name":"Basic and Applied Scientific Research Centre (BASRC), Imam Abdulrahman bin Faisal University","place":["Dammam, Saudi Arabia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ridha","family":"Ben Salah","sequence":"additional","affiliation":[{"name":"Laboratory of Biophysics and Medical Technologies, Higher Institute of Medical Technologies of Tunis (ISTMT), University of Tunis El Manar","place":["Tunis, Tunisia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hanene","family":"Boussi Rahmouni","sequence":"additional","affiliation":[{"name":"Laboratory of Biophysics and Medical Technologies, Higher Institute of Medical Technologies of Tunis (ISTMT), University of Tunis El Manar","place":["Tunis, Tunisia"]},{"name":"The Computer Science Research Centre, University of the West of England","place":["Bristol, United Kingdom"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,7,28]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"1679","DOI":"10.3390\/medicina59091679","article-title":"Advancements in standardizing radiological reports: a comprehensive review","volume":"59","author":"Pesapane","year":"2023","journal-title":"Medicina (B Aires)"},{"key":"B2","doi-asserted-by":"publisher","first-page":"1373420","DOI":"10.3389\/fimag.2024.1373420","article-title":"ChestBioX-Gen: contextual biomedical report generation from chest x-ray images using BioGPT and co-attention mechanism","volume":"3","author":"Ouis","year":"2024","journal-title":"Front Imaging"},{"key":"B3","doi-asserted-by":"publisher","first-page":"715","DOI":"10.1016\/j.jacr.2025.05.003","article-title":"Advancing health care delivery in community, rural, and safety-net settings","volume":"22","author":"Miles","year":"2025","journal-title":"J Am Coll Radiol"},{"key":"B4","doi-asserted-by":"publisher","first-page":"956","DOI":"10.2214\/AJR.15.14605","article-title":"After-hours radiology: challenges and strategies for the radiologist","volume":"205","author":"Rohatgi","year":"2015","journal-title":"Am J Roentgenol"},{"key":"B5","doi-asserted-by":"publisher","first-page":"5199","DOI":"10.1038\/s41598-024-55792-1","article-title":"Real-world testing of an artificial intelligence algorithm for the analysis of chest x-rays in primary care settings","volume":"14","author":"Mir\u00f3 Catalina","year":"2024","journal-title":"Sci Rep"},{"key":"B6","doi-asserted-by":"publisher","first-page":"1292","DOI":"10.1016\/j.injury.2022.12.001","article-title":"The sensitivity of chest x-ray (CXR) for the detection of significant thoracic injury in children exposed to blast","volume":"54","author":"Sargent","year":"2023","journal-title":"Injury"},{"key":"B7","doi-asserted-by":"publisher","first-page":"995","DOI":"10.1016\/j.resinv.2024.08.014","article-title":"Lung imaging in COPD and asthma","volume":"62","author":"Tanabe","year":"2024","journal-title":"Respir Investig"},{"key":"B8","doi-asserted-by":"publisher","first-page":"2211","DOI":"10.3390\/tomography9060171","article-title":"Chest x-ray at emergency admission and potential association with barotrauma in mechanically ventilated patients: experience from the Italian core of the first pandemic peak","volume":"9","author":"Bonaffini","year":"2023","journal-title":"Tomography"},{"key":"B9","doi-asserted-by":"publisher","first-page":"1443","DOI":"10.12785\/ijcds\/1501102","article-title":"Automatic detection of COVID-19 from chest x-ray images using EfficientNet-B7 CNN model with channel-wise attention","volume":"15","author":"Naidji","year":"2024","journal-title":"Int J Com Dig Sys"},{"key":"B10","doi-asserted-by":"publisher","first-page":"7255","DOI":"10.1007\/s00330-024-10794-5","article-title":"Performance of AI to exclude normal chest radiographs to reduce radiologists' Workload","volume":"34","author":"Schalekamp","year":"2024","journal-title":"Eur Radiol"},{"key":"B11","doi-asserted-by":"publisher","first-page":"100287","DOI":"10.1101\/2025.03.18.25324193","article-title":"\u201cLarge language models in radiology reporting\u2014a systematic review of performance, limitations, and clinical implications\u201d","volume":"12","author":"Artsi","year":"2025"},{"key":"B12","doi-asserted-by":"publisher","first-page":"1303","DOI":"10.3390\/bioengineering12121303","article-title":"Agentic AI and large language models in radiology: opportunities and hallucination challenges","volume":"12","author":"Salehi","year":"2025","journal-title":"Bioengineering"},{"key":"B13","doi-asserted-by":"publisher","first-page":"106227","DOI":"10.1016\/j.ijmedinf.2025.106227","article-title":"Vision-language models in diagnostic imaging: review of technical advances, clinical validation, and practical deployment","volume":"208","author":"Dutta","year":"2026","journal-title":"Int J Med Inf"},{"key":"B14","doi-asserted-by":"publisher","first-page":"e32690","DOI":"10.2196\/32690","article-title":"Vision-language model for generating textual descriptions from clinical images: model development and validation study","volume":"8","author":"Ji","year":"2024","journal-title":"JMIR Form Res"},{"key":"B15","first-page":"283","article-title":"SERPENT-VLM: self-refining radiology report generation using vision language models","author":"Kapadnis","year":"2024"},{"key":"B16","doi-asserted-by":"publisher","first-page":"344","DOI":"10.1007\/s10462-025-11337-0","article-title":"Automatic radiology report generation with deep learning: a comprehensive review of methods and advances","volume":"58","author":"Li","year":"2025","journal-title":"Artif Intell Rev"},{"key":"B17","doi-asserted-by":"publisher","first-page":"4171","DOI":"10.1038\/s41598-023-31223-5","article-title":"Medical image captioning via generative pretrained transformers","volume":"13","author":"Selivanov","year":"2023","journal-title":"Sci Rep"},{"key":"B18","first-page":"2863","article-title":"Unify, align and refine: multi-level semantic alignment for radiology report generation","author":"Li","year":"2023"},{"key":"B19","doi-asserted-by":"publisher","first-page":"106251","DOI":"10.1016\/j.bspc.2024.106251","article-title":"A medical report generation method integrating teacher\u2013student model and encoder\u2013decoder network","volume":"94","author":"Zhang","year":"2024","journal-title":"Biomed Signal Process Control"},{"key":"B20","first-page":"1439","article-title":"Generating radiology reports via memory-driven transformer","author":"Chen","year":"2020"},{"key":"B21","author":"Wang","year":"2024"},{"key":"B22","doi-asserted-by":"publisher","first-page":"102814","DOI":"10.1016\/j.artmed.2024.102814","article-title":"Automated image label extraction from radiology reports \u2014 a review","volume":"149","author":"Pereira","year":"2024","journal-title":"Artif Intell Med"},{"key":"B23","doi-asserted-by":"publisher","first-page":"30","DOI":"10.3390\/bioengineering13010030","article-title":"Accurate clinical entity recognition and code mapping of anatomopathological reports using BioClinicalBERT enhanced by retrieval-augmented generation: a hybrid deep learning approach","volume":"13","author":"Abdaoui","year":"2026","journal-title":"Bioengineering"},{"key":"B24","doi-asserted-by":"publisher","first-page":"124644","DOI":"10.1016\/j.eswa.2024.124644","article-title":"CheXReport: a transformer-based architecture to generate chest x-ray reports suggestions","volume":"255","author":"Zeiser","year":"2024","journal-title":"Expert Syst Appl"},{"key":"B25","doi-asserted-by":"publisher","first-page":"5904","DOI":"10.18653\/v1\/2021.acl-long.459","article-title":"Cross-modal memory networks for radiology report generation","volume":"1","author":"Hen","year":"2021"},{"key":"B26","doi-asserted-by":"publisher","first-page":"157651","DOI":"10.1109\/ACCESS.2025.3606961","article-title":"Chest x-ray report generation using abnormality guided vision language model","volume":"13","author":"Edirisinghe","year":"2025","journal-title":"IEEE Access"},{"key":"B27","first-page":"199","article-title":"LLM-Driven Chest x-ray report generation with a modular, reduced-size architecture","author":"Viana Vargas","year":"2024"},{"key":"B28","doi-asserted-by":"publisher","first-page":"295","DOI":"10.1002\/ird3.115","article-title":"\u201cNatural language processing for chest X-ray reports in the transformer era: bERT-like encoders for comprehension and GPT-like decoders for generation,\u201d","volume":"4","author":"Yuan","year":"2025","journal-title":"iRADIOLOGY"},{"key":"B29","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52734.2025.02867","article-title":"FactCheXcker: mitigating measurement hallucinations in chest X-ray report generation models","author":"Heiman","year":"2025"},{"key":"B30","doi-asserted-by":"publisher","first-page":"28191","DOI":"10.1007\/s00521-025-11666-9","article-title":"A survey on retrieval-augmentation generation (RAG) models for healthcare applications","volume":"37","author":"Abo El-Enen","year":"2025","journal-title":"Neural Comput Appl"},{"key":"B31","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP49660.2025.10889152","article-title":"CoMT: chain-of-medical-thought reduces hallucination in medical report generation","author":"Jiang","year":"2025"},{"key":"B32","doi-asserted-by":"crossref","first-page":"10974","DOI":"10.3390\/app152010974","article-title":"Graph-Driven medical report generation with adaptive knowledge distillation","volume":"15","author":"Chen","year":"2025","journal-title":"Appl Sci"},{"key":"B33","doi-asserted-by":"publisher","first-page":"637","DOI":"10.1007\/s11227-025-07111-2","article-title":"Optimizing medical image report generation through a discrete diffusion framework","volume":"81","author":"Sun","year":"2025","journal-title":"J Supercomput"},{"key":"B34","first-page":"33","article-title":"MedPAO: a protocol-driven agent for structuring medical reports","author":"Vaidya","year":"2025"},{"key":"B35","author":"Wang","year":"2025"},{"key":"B36","author":"Loshchilov","year":"2017"},{"key":"B37","author":"Zhang","year":"2019"},{"key":"B38","first-page":"311","article-title":"Bleu: a method for automatic evaluation of machine translation","author":"Papineni","year":"2002"},{"key":"B39","first-page":"74","article-title":"ROUGE: a package for automatic evaluation of summaries","author":"Lin","year":"2004"},{"key":"B40","first-page":"65","article-title":"METEOR: an automatic metric for MT evaluation with improved correlation with human judgments","author":"Banerjee","year":"2005"},{"key":"B41","doi-asserted-by":"publisher","first-page":"590","DOI":"10.1609\/aaai.v33i01.3301590","article-title":"CheXpert: a large chest radiograph dataset with uncertainty labels and expert comparison","volume":"33","author":"Irvin","year":"2019","journal-title":"Proceedings of the AAAI Conference on Artificial Intelligence"},{"key":"B42","article-title":"RadGraph: extracting clinical entities and relations from radiology reports","author":"Jain","year":"2021"},{"key":"B43","first-page":"249","article-title":"Clinically accurate chest X-ray report generation","author":"Liu","year":"2019"},{"key":"B44","first-page":"5288","article-title":"Improving factual completeness and consistency of image-to-text radiology report generation","author":"Miura","year":"2021"},{"key":"B45","first-page":"15004","article-title":"Ratescore: a metric for radiology report generation","author":"Zhao","year":"2024"},{"key":"B46","article-title":"ReXrank: a public leaderboard for AI-powered radiology report generation","author":"Zhang","year":"2024"},{"key":"B47","doi-asserted-by":"publisher","first-page":"100802","DOI":"10.1016\/j.patter.2023.100802","article-title":"Evaluating progress in automatic chest X-ray radiology report generation","volume":"4","author":"Yu","year":"2023","journal-title":"Patterns"},{"key":"B48","article-title":"Cross-modal memory networks for radiology report generation","author":"Chen","year":"2022"},{"key":"B49","author":"Yi","year":"2025"},{"key":"B50","first-page":"1081","article-title":"Rule: reliable multimodal RAG for factuality in medical vision language models","author":"Xia","year":"2024"},{"key":"B51","unstructured":"MDocAgent: a multi-modal multi-agent framework for document understanding\n          \n          \n            \n              Han\n              S\n            \n            \n              Xia\n              P\n            \n            \n              Zhang\n              R\n            \n            \n              Sun\n              T\n            \n            \n              Li\n              Y\n            \n            \n              Zhu\n              H\n            \n          \n          \n          2025"},{"key":"B52","author":"Qi","year":"2026"},{"key":"B53","first-page":"1","article-title":"Mrgagents: a multi-agent framework for improved medical report generation with med-lvlms","author":"Wang","year":"2025"},{"key":"B54","doi-asserted-by":"publisher","first-page":"317","DOI":"10.1038\/s41597-019-0322-0","article-title":"MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports","volume":"6","author":"Johnson","year":"2019","journal-title":"Sci Data"},{"key":"B55","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2405.19538","article-title":"CheXpert plus: augmenting a large chest X-ray dataset with text radiology reports, patient demographics and additional image formats","author":"Chambon","year":"2024"}],"container-title":["Frontiers in Radiology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fradi.2026.1875905\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,28]],"date-time":"2026-07-28T11:10:19Z","timestamp":1785237019000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fradi.2026.1875905\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,28]]},"references-count":55,"alternative-id":["10.3389\/fradi.2026.1875905"],"URL":"https:\/\/doi.org\/10.3389\/fradi.2026.1875905","relation":{},"ISSN":["2673-8740"],"issn-type":[{"value":"2673-8740","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,7,28]]},"article-number":"1875905"}}