{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T04:50:21Z","timestamp":1791521421069,"version":"4.3.4"},"reference-count":25,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T00:00:00Z","timestamp":1791504000000},"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. Cardiovasc. Med."],"abstract":"<jats:sec>\n                    <jats:title>Objective<\/jats:title>\n                    <jats:p>To compare ChatGPT-4o, ChatGPT-5, and DeepSeek-R1 in assigning main Coronary Artery Disease Reporting and Data System (CAD-RADS) categories from findings-only coronary computed tomography angiography (CCTA) reports and to characterize management-relevant classification errors.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>This retrospective diagnostic-accuracy study included 2,100 category-balanced findings-only CCTA reports (300 per CAD-RADS category). Two experienced CCTA radiologists graded the English-translated reports; 12 disagreements were resolved by a third. ChatGPT-4o was evaluated in standard mode, ChatGPT-5 in Thinking mode, and DeepSeek-R1 in Deep Thinking mode; model generation and inference mode therefore varied together. Exact-match accuracy with 95% Wilson confidence intervals (CIs), paired comparisons, and unweighted Gwet AC1 with 95% CIs were calculated. Management-relevant transitions and digital subtraction angiography (DSA) findings among high-risk downgrades were analyzed.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      DeepSeek-R1 had an exact-match accuracy of 82.57% (95% CI, 80.89\u201384.13) and AC1 of 0.797 (95% CI, 0.778\u20130.816). ChatGPT-5 had an accuracy of 80.19% (95% CI, 78.43\u201381.84) and AC1 of 0.769 (95% CI, 0.749\u20130.789), whereas ChatGPT-4o had an accuracy of 53.62% (95% CI, 51.48\u201355.74) and AC1 of 0.460 (95% CI, 0.435\u20130.485). Cochran's\n                      <jats:italic>Q<\/jats:italic>\n                      test was significant (Q\u2009=\u2009657.87; df\u2009=\u20092;\n                      <jats:italic>P<\/jats:italic>\n                      \u2009&amp;lt;\u20090.001); after Holm correction, DeepSeek-R1 also outperformed ChatGPT-5 in the paired comparison (\n                      <jats:italic>P<\/jats:italic>\n                      \u2009=\u20090.026), and both outperformed ChatGPT-4o (both\n                      <jats:italic>P<\/jats:italic>\n                      \u2009&amp;lt;\u20090.001). ChatGPT-4o downgraded 127\/900 (14.11%) reference high-risk reports to CAD-RADS 3 or lower. Of these 127 cases, 65 underwent DSA within three months; DSA supported the original high-risk assessment in 56\/65 (86.15%). The corresponding high-risk downgrade proportions were 37\/900 (4.11%) for ChatGPT-5 and 23\/900 (2.56%) for DeepSeek-R1.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>The DeepSeek-R1 Deep Thinking and ChatGPT-5 Thinking configurations achieved higher accuracy than ChatGPT-4o in standard mode. Because model generation and inference mode varied together, the results do not isolate a model-family effect. LLMs merit further evaluation for secondary structuring when no documented CAD-RADS category is available, but should not overwrite a radiologist-assigned category or independently determine management.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fcvm.2026.1909068","type":"journal-article","created":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T04:25:51Z","timestamp":1791519951000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Large language model-assisted CAD-RADS classification: structured report extraction, accuracy, and clinical decision-making risk of ChatGPT-4o\/5 and DeepSeek-R1"],"prefix":"10.3389","volume":"13","author":[{"given":"Haorong","family":"Rong","sequence":"first","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiyi","family":"Huang","sequence":"additional","affiliation":[{"name":"Department of Clinical Laboratory, Lecong Hospital of Shunde","place":["Foshan, Guangdong, China"]},{"name":"Department of Clinical Laboratory, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xinjie","family":"Chen","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lizhu","family":"Ouyang","sequence":"additional","affiliation":[{"name":"Department of Ultrasound, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yuelu","family":"Li","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xin","family":"Lin","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jialing","family":"Pan","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yunjing","family":"Li","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fan","family":"Xie","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fusheng","family":"Ouyang","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qiugen","family":"Hu","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Baoliang","family":"Guo","sequence":"additional","affiliation":[{"name":"Department of Radiology, The Eighth Affiliated Hospital, Southern Medical University","place":["Foshan, Guangdong, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,10,9]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"1974","DOI":"10.1016\/j.jcmg.2022.07.002","article-title":"CAD-RADS 2.0-2022 coronary artery disease-reporting and data system: an expert consensus document","volume":"15","author":"Cury","year":"2022","journal-title":"JACC Cardiovasc Imaging"},{"key":"B2","doi-asserted-by":"publisher","first-page":"55","DOI":"10.1007\/s10554-024-03283-9","article-title":"Assessing the need for coronary angiography in high-risk non-ST-elevation acute coronary syndrome patients using artificial intelligence and computed tomography","volume":"41","author":"Cagnina","year":"2025","journal-title":"Int J Cardiovasc Imaging"},{"key":"B3","doi-asserted-by":"publisher","first-page":"80","DOI":"10.4274\/dir.2023.232417","article-title":"Large language models in radiology: fundamentals, applications, ethical considerations, risks, and future directions","volume":"30","author":"Akinci D\u2019Antonoli","year":"2024","journal-title":"Diagn Interv Radiol"},{"key":"B4","doi-asserted-by":"publisher","first-page":"e076865","DOI":"10.1136\/bmjopen-2023-076865","article-title":"Large language model-based information extraction from free-text radiology reports: a scoping review protocol","volume":"13","author":"Reichenpfader","year":"2023","journal-title":"BMJ Open"},{"key":"B5","doi-asserted-by":"publisher","first-page":"e67706","DOI":"10.2196\/67706","article-title":"Extracting pulmonary embolism diagnoses from radiology impressions using GPT-4o: large language model evaluation study","volume":"13","author":"Mahyoub","year":"2025","journal-title":"JMIR Med Inform"},{"key":"B6","doi-asserted-by":"publisher","first-page":"e65984","DOI":"10.2196\/65984","article-title":"Large language model applications for health information extraction in oncology: scoping review","volume":"11","author":"Chen","year":"2025","journal-title":"JMIR Cancer"},{"key":"B7","doi-asserted-by":"publisher","first-page":"112341","DOI":"10.1016\/j.ejrad.2025.112341","article-title":"Evaluation of GPT-4o for multilingual translation of radiology reports across imaging modalities","volume":"191","author":"Terzis","year":"2025","journal-title":"Eur J Radiol"},{"key":"B8","doi-asserted-by":"publisher","first-page":"112316","DOI":"10.1016\/j.ejrad.2025.112316","article-title":"Comparing large language models and text embedding models for automated classification of textual, semantic, and critical changes in radiology reports","volume":"191","author":"Lindholz","year":"2025","journal-title":"Eur J Radiol"},{"key":"B9","doi-asserted-by":"publisher","first-page":"2589","DOI":"10.1007\/s00330-024-11107-6","article-title":"Large language models for structured reporting in radiology: past, present, and future","volume":"35","author":"Busch","year":"2025","journal-title":"Eur Radiol"},{"key":"B10","doi-asserted-by":"publisher","first-page":"e241073","DOI":"10.1148\/radiol.241073","article-title":"Open-source large language models in radiology: a review and tutorial for practical research and clinical deployment","volume":"314","author":"Savage","year":"2025","journal-title":"Radiology"},{"key":"B11","doi-asserted-by":"publisher","first-page":"60","DOI":"10.1038\/s41591-024-03425-5","article-title":"The TRIPOD-LLM reporting guideline for studies using large language models","volume":"31","author":"Gallifant","year":"2025","journal-title":"Nat Med"},{"key":"B12","doi-asserted-by":"publisher","first-page":"1123","DOI":"10.3348\/kjr.2025.1522","article-title":"Minimum reporting items for clear evaluation of accuracy reports of large language models in healthcare (MI-CLEAR-LLM): 2025 updates","volume":"26","author":"Park","year":"2025","journal-title":"Korean J Radiol"},{"key":"B13","doi-asserted-by":"publisher","first-page":"e232133","DOI":"10.1148\/radiol.232133","article-title":"BI-RADS category assignments by GPT-3.5, GPT-4, and google bard: a multilanguage study","volume":"311","author":"Cozzi","year":"2024","journal-title":"Radiology"},{"key":"B14","doi-asserted-by":"publisher","first-page":"1578","DOI":"10.1111\/liv.15891","article-title":"Using GPT-4 for LI-RADS feature extraction and categorization with multilingual free-text reports","volume":"44","author":"Gu","year":"2024","journal-title":"Liver Int"},{"key":"B15","doi-asserted-by":"publisher","first-page":"110455","DOI":"10.1016\/j.clinimag.2025.110455","article-title":"ChatGPT vs gemini: comparative accuracy and efficiency in lung-RADS score assignment from radiology reports","volume":"121","author":"Singh","year":"2025","journal-title":"Clin Imaging"},{"key":"B16","doi-asserted-by":"publisher","first-page":"368","DOI":"10.1093\/bjr\/tqae236","article-title":"Assessing the performance of ChatGPT and bard\/gemini against radiologists for prostate imaging reporting and data system classification based on prostate multiparametric MRI text reports","volume":"98","author":"Lee","year":"2025","journal-title":"Br J Radiol"},{"key":"B17","doi-asserted-by":"publisher","first-page":"2303","DOI":"10.1007\/s10278-024-01328-y","article-title":"ChatGPT vs gemini: comparative accuracy and efficiency in CAD-RADS score assignment from radiology reports","volume":"38","author":"Silbergleit","year":"2025","journal-title":"J Imaging Inform Med"},{"key":"B18","doi-asserted-by":"publisher","first-page":"322","DOI":"10.1016\/j.jcct.2025.03.007","article-title":"Performance of large language models for CAD-RADS 2.0 classification derived from cardiac CT reports","volume":"19","author":"Arnold","year":"2025","journal-title":"J Cardiovasc Comput Tomogr"},{"key":"B19","doi-asserted-by":"publisher","first-page":"817","DOI":"10.3348\/kjr.2025.0293","article-title":"Large language models for CAD-RADS 2.0 extraction from semi-structured coronary CT angiography reports: a multi-institutional study","volume":"26","author":"Min","year":"2025","journal-title":"Korean J Radiol"},{"key":"B20","doi-asserted-by":"publisher","first-page":"217","DOI":"10.1097\/RCT.0000000000001802","article-title":"Enhancing the CAD-RADS 2.0 category assignment performance of ChatGPT and DeepSeek through few-shot prompting","volume":"50","author":"Kaya","year":"2026","journal-title":"J Comput Assist Tomogr"},{"key":"B21","doi-asserted-by":"publisher","first-page":"112","DOI":"10.1186\/s13244-026-02285-6","article-title":"Comparative evaluation of large language models for generating CAD-RADS 2.0-compliant diagnostic conclusions in cardiac CT reports","volume":"17","author":"Lorusso","year":"2026","journal-title":"Insights Imaging"},{"key":"B22","doi-asserted-by":"publisher","first-page":"61","DOI":"10.1186\/1471-2288-13-61","article-title":"A comparison of Cohen\u2019s kappa and gwet\u2019s AC1 when calculating inter-rater reliability coefficients: a study conducted with personality disorder samples","volume":"13","author":"Wongpakaran","year":"2013","journal-title":"BMC Med Res Methodol"},{"key":"B23","volume-title":"R: A Language and Environment for Statistical Computing","year":"2026"},{"key":"B24","doi-asserted-by":"publisher","first-page":"633","DOI":"10.1038\/s41586-025-09422-z","article-title":"DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning","volume":"645","author":"Guo","year":"2025","journal-title":"Nature"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.1007\/s00117-026-01580-z","article-title":"Retrieval-augmented generation-enhanced large language models for comprehensive CAD-RADS 2.0 categorization from structured coronary CTA reports","author":"Kaba","year":"2026","journal-title":"Radiologie (Heidelb)"}],"container-title":["Frontiers in Cardiovascular Medicine"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fcvm.2026.1909068\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T04:25:52Z","timestamp":1791519952000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fcvm.2026.1909068\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,9]]},"references-count":25,"alternative-id":["10.3389\/fcvm.2026.1909068"],"URL":"https:\/\/doi.org\/10.3389\/fcvm.2026.1909068","relation":{},"ISSN":["2297-055X"],"issn-type":[{"value":"2297-055X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,9]]},"article-number":"1909068"}}