{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,28]],"date-time":"2026-09-28T11:55:50Z","timestamp":1790596550735,"version":"4.1.0"},"reference-count":40,"publisher":"JMIR Publications Inc.","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["JMIR AI"],"abstract":"Abstract<\/jats:title>\n \n Background<\/jats:title>\n Recent progress has demonstrated the potential of deep learning models in analyzing electrocardiogram (ECG) pathologies. However, this method is intricate, expensive to develop, and designed for specific purposes. Large language models show promise in medical image interpretation, and yet their effectiveness in ECG analysis remains understudied. Generative Pretrained Transformer 4 Omni (GPT-4o), a multimodal artificial intelligence model, capable of processing images and text without task-specific training, may offer an accessible alternative.<\/jats:p>\n <\/jats:sec>\n \n Objective<\/jats:title>\n This study aimed to evaluate GPT-4o\u2019s effectiveness in interpreting 12-lead ECGs, assessing classification accuracy, and exploring methods to enhance its performance.<\/jats:p>\n <\/jats:sec>\n \n Methods<\/jats:title>\n A total of 6 common ECG diagnoses were evaluated: normal ECG, ST-segment elevation myocardial infarction, atrial fibrillation, right bundle branch block, left bundle branch block, and paced rhythm, with 30 normal ECGs and 10 of each abnormal pattern, totaling 80 cases. Deidentified ECGs were analyzed using OpenAI\u2019s GPT-4o. Our study used both zero-shot and few-shot learning methodologies to investigate three main scenarios: (1) ECG image recognition, (2) binary classification of normal versus abnormal ECGs, and (3) multiclass classification into 6 categories.<\/jats:p>\n <\/jats:sec>\n \n Results<\/jats:title>\n The model excelled in recognizing ECG images, achieving an accuracy of 100%. In the classification of normal or abnormal ECG cases, the few-shot learning approach improved GPT-4o\u2019s accuracy by 30% from the baseline, reaching 83% (95% CI 81.8%-84.6%). However, multiclass classification for a specific pathology remained limited, achieving only 41% accuracy.<\/jats:p>\n <\/jats:sec>\n \n Conclusions<\/jats:title>\n GPT-4o effectively differentiates normal from abnormal ECGs, suggesting its potential as an accessible artificial intelligence\u2013assisted triage tool. Although limited in diagnosing specific cardiac conditions, GPT-4o\u2019s capability to interpret ECG images without specialized training highlights its potential for preliminary ECG interpretation in clinical and remote settings.<\/jats:p>\n <\/jats:sec>","DOI":"10.2196\/74426","type":"journal-article","created":{"date-parts":[[2025,7,9]],"date-time":"2025-07-09T15:50:08Z","timestamp":1752076208000},"page":"e74426-e74426","source":"Crossref","is-referenced-by-count":7,"title":["Effectiveness of the GPT-4o Model in Interpreting Electrocardiogram Images for Cardiac Diagnostics: Diagnostic Accuracy Study"],"prefix":"10.2196","volume":"4","author":[{"ORCID":"https:\/\/orcid.org\/0009-0005-6721-5158","authenticated-orcid":false,"given":"Haya","family":"Engelstein","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2529-1567","authenticated-orcid":false,"given":"Roni","family":"Ramon-Gonen","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4295-6679","authenticated-orcid":false,"given":"Avi","family":"Sabbag","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4567-3108","authenticated-orcid":false,"given":"Eyal","family":"Klang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-5118-8619","authenticated-orcid":false,"given":"Karin","family":"Sudri","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0007-6532-9252","authenticated-orcid":false,"given":"Michal","family":"Cohen-Shelly","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7416-5989","authenticated-orcid":false,"given":"Israel","family":"Barbash","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1010","published-online":{"date-parts":[[2025,8,22]]},"reference":[{"issue":"3","key":"R1","doi-asserted-by":"publisher","first-page":"279","DOI":"10.1093\/ehjdh\/ztad029","article-title":"ChatGPT takes on the European exam in core cardiology: an artificial intelligence success story?","volume":"4","author":"Skalidis","journal-title":"Eur Heart J Digit Health"},{"issue":"2","key":"R2","doi-asserted-by":"publisher","first-page":"100","DOI":"10.1038\/s41569-018-0104-y","article-title":"Computational models in cardiology","volume":"16","author":"Niederer","journal-title":"Nat Rev Cardiol"},{"issue":"1","key":"R3","doi-asserted-by":"publisher","DOI":"10.1098\/rsfs.2019.0122","article-title":"Deep medical image analysis with representation learning and neuromorphic computing","volume":"11","author":"Getty","journal-title":"Interface Focus"},{"issue":"3","key":"R4","doi-asserted-by":"publisher","DOI":"10.1088\/1361-6579\/aaaa9d","article-title":"A robust deep convolutional neural network for the classification of abnormal cardiac rhythm using single lead electrocardiograms of variable length","volume":"39","author":"Kamaleswaran","journal-title":"Physiol Meas"},{"key":"R5","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijmedinf.2024.105753","article-title":"A systematic review on the impact of artificial intelligence on electrocardiograms in cardiology","volume":"195","author":"Oke","journal-title":"Int J Med Inform"},{"issue":"5","key":"R6","doi-asserted-by":"publisher","first-page":"428","DOI":"10.1001\/jamacardio.2019.0640","article-title":"Development and validation of a deep-learning model to screen for hyperkalemia from the electrocardiogram","volume":"4","author":"Galloway","journal-title":"JAMA Cardiol"},{"issue":"38","key":"R7","doi-asserted-by":"publisher","first-page":"3948","DOI":"10.1093\/eurheartj\/ehab588","article-title":"Deep learning analysis of electrocardiogram for risk prediction of drug-induced arrhythmias and diagnosis of long QT syndrome","volume":"42","author":"Prifti","journal-title":"Eur Heart J"},{"issue":"30","key":"R8","doi-asserted-by":"publisher","first-page":"2885","DOI":"10.1093\/eurheartj\/ehab153","article-title":"Electrocardiogram screening for aortic valve stenosis using artificial intelligence","volume":"42","author":"Cohen-Shelly","journal-title":"Eur Heart J"},{"issue":"9","key":"R9","doi-asserted-by":"publisher","DOI":"10.1161\/CIRCEP.119.007284","article-title":"Age and sex estimation using artificial intelligence from standard 12-lead ECGs","volume":"12","author":"Attia","journal-title":"Circ Arrhythm Electrophysiol"},{"key":"R10","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijmedinf.2022.104790","article-title":"Short-term atrial fibrillation detection using electrocardiograms: a comparison of machine learning approaches","volume":"163","author":"Jahan","journal-title":"Int J Med Inform"},{"key":"R11","doi-asserted-by":"crossref","unstructured":"Bochinski E Eiselein V Sikora T . Training a convolutional neural network for multi-class object detection using solely virtual world data. Presented at: 2016 13th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS); Aug 23-26, 2016; Colorado Springs, CO, USA. [doi: 10.1109\/AVSS.2016.7738056]","DOI":"10.1109\/AVSS.2016.7738056"},{"issue":"7553","key":"R12","doi-asserted-by":"publisher","first-page":"436","DOI":"10.1038\/nature14539","article-title":"Deep learning","volume":"521","author":"LeCun","journal-title":"Nature New Biol"},{"issue":"1","key":"R13","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-024-55158-1","article-title":"Deep neural networks and humans both benefit from compositional language structure","volume":"15","author":"Galke","journal-title":"Nat Commun"},{"key":"R14","doi-asserted-by":"publisher","DOI":"10.1016\/j.jemermed.2024.12.010","article-title":"Can GPT-4o accurately diagnose trauma x-rays? A comparative study with expert evaluations","volume":"73","author":"\u00d6zt\u00fcrk","journal-title":"J Emerg Med"},{"issue":"12","key":"R15","doi-asserted-by":"publisher","DOI":"10.7759\/cureus.75532","article-title":"Evaluating the role of GPT-4 and GPT-4o in the detectability of chest radiography reports requiring further assessment","volume":"16","author":"Kanzawa","journal-title":"Cureus"},{"key":"R16","doi-asserted-by":"publisher","DOI":"10.1016\/j.jelectrocard.2024.153851","article-title":"In the face of confounders: atrial fibrillation detection\u2014practitioners vs. ChatGPT","volume":"88","author":"Avidan","journal-title":"J Electrocardiol"},{"issue":"11","key":"R17","doi-asserted-by":"publisher","DOI":"10.3390\/diagnostics15111320","article-title":"Do LLMs have \u201cthe Eye\u201d for MRI? Evaluating GPT-4o, Grok, and Gemini on brain MRI performance: first evaluation of Grok in medical imaging and a comparative analysis","volume":"15","author":"Sozer","journal-title":"Diagnostics (Basel)"},{"issue":"11","key":"R18","doi-asserted-by":"publisher","first-page":"4365","DOI":"10.1016\/j.acra.2024.09.005","article-title":"Evaluating GPT-4o\u2019s performance in the official European board of radiology exam: a comprehensive assessment","volume":"31","author":"Be\u015fler","journal-title":"Acad Radiol"},{"key":"R19","unstructured":"Hampton J . The ECG Made Easy. 9th ed. Elsevier; 2019. ISBN: 0702074578"},{"key":"R20","doi-asserted-by":"crossref","unstructured":"Kadam S Vaidya V . Review and analysis of zero, one and few shot learning approaches. In: Advances in Intelligent Systems and Computing. Springer, Cham; 2019:100-112. [doi: 10.1007\/978-3-030-16657-1_10]","DOI":"10.1007\/978-3-030-16657-1_10"},{"key":"R21","unstructured":"LastMile AI Docs. AI concepts. 2023. URL: https:\/\/lastmile-ai.gitbook.io\/lastmile-ai-docs\/getting-started\/ai-concepts [Accessed 14-08-2025]"},{"key":"R22","unstructured":"Brown TB Mann B Ryder N Subbiah M Kaplan J Dhariwal P et al. Language models are few-shot learners. arXiv. Preprint posted online on 2020. [doi: 10.48550\/ARXIV.2005.14165]"},{"issue":"4","key":"R23","doi-asserted-by":"publisher","first-page":"1959","DOI":"10.1007\/s00330-024-11035-5","article-title":"Assessing GPT-4 multimodal performance in radiological image analysis","volume":"35","author":"Brin","journal-title":"Eur Radiol"},{"key":"R24","doi-asserted-by":"crossref","unstructured":"Natarajan A Chang Y Mariani S et al. A wide and deep transformer neural network for 12-lead ECG classification. Presented at: 2020 Computing in Cardiology Conference; Sep 13-16, 2020; Rimini, Italy. [doi: 10.22489\/CinC.2020.107]","DOI":"10.22489\/CinC.2020.107"},{"issue":"8","key":"R25","doi-asserted-by":"publisher","first-page":"1179","DOI":"10.1093\/europace\/euaa377","article-title":"Deep learning and the electrocardiogram: review of the current state-of-the-art","volume":"23","author":"Somani","journal-title":"Europace"},{"issue":"3","key":"R26","doi-asserted-by":"publisher","DOI":"10.3390\/jpm12030336","article-title":"Diagnostic accuracy of the deep learning model for the detection of ST elevation myocardial infarction on electrocardiogram","volume":"12","author":"Choi","journal-title":"J Pers Med"},{"issue":"3.3","key":"R27","doi-asserted-by":"publisher","first-page":"401","DOI":"10.14419\/ijet.v7i2.33.14195","article-title":"Classification of cardiac arrhythmias using deep learning","volume":"7","author":"Hwan Kim","journal-title":"IJET"},{"issue":"11","key":"R28","doi-asserted-by":"publisher","first-page":"8755","DOI":"10.1007\/s00521-022-06889-z","article-title":"Automated ECG multi-class classification system based on combining deep learning features with HRV and ECG measures","volume":"34","author":"Eltrass","journal-title":"Neural Comput Appl"},{"key":"R29","doi-asserted-by":"publisher","DOI":"10.1016\/j.imu.2018.08.002","article-title":"Multiclass classification of myocardial infarction with convolutional and recurrent neural networks for portable ECG devices","volume":"13","author":"Lui","journal-title":"Inform Med Unlocked"},{"key":"R30","doi-asserted-by":"crossref","unstructured":"Jin L Dong J . Normal versus abnormal ECG classification by the aid of deep learning. In: Artificial Intelligence\u2014Emerging Trends and Applications. InTechOpen; 2018:295-315. [doi: 10.5772\/intechopen.75546] ISBN: 978-1-78923-364-3","DOI":"10.5772\/intechopen.75546"},{"issue":"4","key":"R31","doi-asserted-by":"publisher","DOI":"10.3390\/e24040471","article-title":"CNN-FWS: a model for the diagnosis of normal and abnormal ECG with feature adaptive","volume":"24","author":"Zhu","journal-title":"Entropy (Basel)"},{"issue":"2","key":"R32","doi-asserted-by":"publisher","first-page":"401","DOI":"10.1007\/s11517-021-02321-1","article-title":"Feasibility study to improve deep learning in OCT diagnosis of rare retinal diseases with few-shot classification","volume":"59","author":"Yoo","journal-title":"Med Biol Eng Comput"},{"key":"R33","doi-asserted-by":"publisher","DOI":"10.1177\/25158414251340569","article-title":"Multimodal LLMs for retinal disease diagnosis via OCT: few-shot versus single-shot learning","volume":"17","author":"Agbareia","journal-title":"Ther Adv Ophthalmol"},{"key":"R34","unstructured":"Zhou G Han Z Chen S Huang B Zhu L Khan S et al. Adapting large multimodal models to distribution shifts: the role of in-context learning. arXiv. Preprint posted online on 2024. [doi: 10.48550\/arXiv.2405.12217]"},{"key":"R35","unstructured":"Han Z Zhou G He R Wang J Wu T Yin Y et al. How well does GPT-4V(ision) adapt to distribution shifts? a preliminary investigation. Preprint posted online on 2024. [doi: 10.48550\/arXiv.2312.07424]"},{"key":"R36","doi-asserted-by":"crossref","unstructured":"Zhang X Li J Chu W Hai J Xu R Yang Y et al. On the out-of-distribution generalization of multimodal large language models. arXiv. Preprint posted online on 2024. [doi: 10.48550\/arXiv.2402.06599]","DOI":"10.1109\/CVPR52734.2025.00965"},{"issue":"1","key":"R37","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-024-58514-9","article-title":"GPT-agents based on medical guidelines can improve the responsiveness and explainability of outcomes for traumatic brain injury rehabilitation","volume":"14","author":"Zhenzhu","journal-title":"Sci Rep"},{"issue":"12","key":"R38","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3571730","article-title":"Survey of hallucination in natural language generation","volume":"55","author":"Ji","journal-title":"ACM Comput Surv"},{"key":"R39","doi-asserted-by":"publisher","DOI":"10.3389\/frai.2023.1278796","article-title":"Performance analysis of large language models in the domain of legal argument mining","volume":"6","author":"Al Zubaer","journal-title":"Front Artif Intell"},{"key":"R40","doi-asserted-by":"publisher","DOI":"10.1016\/j.artmed.2024.102949","article-title":"A systematic review of few-shot learning in medical imaging","volume":"156","author":"Pachetti","journal-title":"Artif Intell Med"}],"container-title":["JMIR AI"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/ai.jmir.org\/2025\/1\/e74426","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T18:05:38Z","timestamp":1782842738000},"score":1,"resource":{"primary":{"URL":"https:\/\/ai.jmir.org\/2025\/1\/e74426"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,8,22]]},"references-count":40,"URL":"https:\/\/doi.org\/10.2196\/74426","relation":{},"ISSN":["2817-1705"],"issn-type":[{"value":"2817-1705","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,8,22]]},"article-number":"v4i3e74426"}}