Performance of Multimodal GPT-4V on USMLE with Image: Potential for Imaging Diagnostic Support with Explanations
Open accessThe DOI matches its Crossref record; 25 of the 25 references it cites were found in a registry or an index.
Abstract
Background: Using artificial intelligence (AI) to help clinical diagnoses has been an active research topic for more than six decades. Past research, however, has not had the scale and accuracy for use in clinical decision making. The power of AI in large language model (LLM)-related technologies may be changing this. In this study, we evaluated the performance and interpretability of Generative Pre-trained Transformer 4 Vision (GPT-4V), a multimodal LLM, on medical licensing examination questions with images. Methods: We used three sets of multiple-choice questions with images from the United States Medical Licensing Examination (USMLE), the USMLE question bank for medical students with different difficulty level (AMBOSS), and the Diagnostic Radiology Qualifying Core Exam (DRQCE) to test GPT-4V’s accuracy and explanation quality. We compared GPT-4V with two state-of-the-art LLMs, GPT-4 and ChatGPT. We also assessed the preference and feedback of healthcare professionals on GPT-4V’s explanations. We presented a case scenario on how GPT-4V can be used for clinical decision support. Results: GPT-4V outperformed ChatGPT (58.4%) and GPT4 (83.6%) to pass the full USMLE exam with an overall accuracy of 90.7%. In comparison, the passing threshold was 60% for medical students. For questions with images, GPT-4V achieved a performance that was equivalent to the 70th - 80th percentile with AMBOSS medical students, with accuracies of 86.2%, 73.1%, and 62.0% on USMLE, DRQCE, and AMBOSS, respectively. While the accuracies decreased quickly among medical students when the difficulties of questions increased, the performance of GPT-4V remained relatively stable. On the other hand, GPT-4V’s performance varied across different medical subdomains, with the highest accuracy in immunology (100%) and otolaryngology (100%) and the lowest accuracy in anatomy (25%) and emergency medicine (25%). When GPT-4V answered correctly, its explanations were almost as good as those made by domain experts. However, when GPT-4V answered incorrectly, the quality of generated explanation was poor: 18.2% wrong answers had made-up text; 45.5% had inferencing errors; and 76.3% had image misunderstandings. Our results show that after experts gave GPT-4V a short hint about the image, it reduced 40.5% errors on average, and more difficult test questions had higher performance gains. Therefore, a hypothetical clinical decision support system as shown in our case scenario is a human-AI-in-the-loop system where a clinician can interact with GPT-4V with hints to maximize its clinical use. Conclusion: GPT-4V outperformed other LLMs and typical medical student performance on results for medical licensing examination questions with images. However, uneven subdomain performance and inconsistent explanation quality may restrict its practical application in clinical settings. The observation that physicians’ hints significantly improved GPT-4V’s performance suggests that future research could focus on developing more effective human-AI collaborative systems. Such systems could potentially overcome current limitations and make GPT-4V more suitable for clinical use. 1-2 sentence description: In this study the authors show that GPT-4V, a large multimodal chatbot, achieved accuracy on medical licensing exams with images equivalent to the 70th - 80th percentile with AMBOSS medical students. The authors also show issues with GPT-4V, including uneven performance in different clinical subdomains and explanation quality, which may hamper its clinical use.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025 · cited 1×
“48 We found that more than 18% of GPT-4V explanations contain hallucinations.”
BackgroundSurvey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y., Chen, D., Dai, W., et al. (2022) · arXiv · doi:10.48550/arxiv.2202.03629 · Cited by 4,981
Kung, T. H., Cheatham, M., Medenilla, A., Sillos, C., De Leon, L., Elepaño, C., Madriaga, M., Aggabao, R., Diaz-Candido, G., Maningo, J., et al. (2023) · PLOS Digital Health · doi:10.1371/journal.pdig.0000198 · Cited by 4,115 · cited 1×
“17 ChatGPT has achieved commendable results in the United States Medical Licensing Examinations, leading to discussions about the readiness of LLM applications for integration into clinical 18–20 , and educational 21–23 environments.”
BackgroundKung, T. H., Cheatham, M., ChatGPT, Medenilla, A., Sillos, C., De Leon, L., Elepaño, C., Madriaga, M., Aggabao, R., Diaz-Candido, G., et al. (2022) · doi:10.1101/2022.12.19.22283643 · Cited by 4,115
Sallam, M. (2023) · Healthcare · doi:10.3390/healthcare11060887 · Cited by 3,126
An overview of clinical decision support systems: Benefits, risks, and strategies for success
Sutton, R. T., Pincock, D., Baumgart, D. C., Sadowski, D. C., Fedorak, R. N., & Kroeker, K. I. (2020) · npj Digital Medicine · doi:10.1038/s41746-020-0221-y · Cited by 3,019
Ayers, J. W., Poliak, A., Dredze, M., Leas, E. C., Zhu, Z., Kelley, J. B., Faix, D. J., Goodman, A. M., Longhurst, C. A., Hogarth, M., et al. (2023) · JAMA Internal Medicine · doi:10.1001/jamainternmed.2023.1838 · Cited by 2,613
Gilson, A., Safranek, C. W., Huang, T., Socrates, V., Chi, L., Taylor, R. A., & Chartash, D. (2023) · JMIR Medical Education · doi:10.2196/45312 · Cited by 2,078
ChatGPT and other large language models are double-edged swords
Shen, Y., Heacock, L., Elias, J., Hentel, K. D., Reig, B., Shih, G., & Moy, L. (2023) · Radiology · doi:10.1148/radiol.230163 · Cited by 1,008 · cited 1×
“16”
Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis
Aggarwal, R., Sounderajah, V., Martin, G., Ting, D. S. W., Karthikesalingam, A., King, D., Ashrafian, H., & Darzi, A. (2021) · npj Digital Medicine · doi:10.1038/s41746-021-00438-z · Cited by 946
Accuracy and Reliability of Chatbot Responses to Physician Questions
Goodman, R. S., Patrinely, J. R., Stone, C. A., Zimmerman, E., Donald, R. R., Chang, S. S., Berkowitz, S. T., Finn, A. P., Jahangir, E., Scoville, E. A., et al. (2023) · JAMA Network Open · doi:10.1001/jamanetworkopen.2023.36483 · Cited by 484 · cited 1×
“17 ChatGPT has achieved commendable results in the United States Medical Licensing Examinations, leading to discussions about the readiness of LLM applications for integration into clinical 18–20 , and educational 21–23 environments.”
BackgroundTowards Generalist Biomedical AI
Tu, T., Azizi, S., Driess, D., Schaekermann, M., Amin, M., Chang, P.-C., Carroll, A., Lau, C., Tanno, R., Ktena, I., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.14334 · Cited by 483 · cited 1×
“26,27 OpenAI's recent introduction of GPT-4V has undeniably made strides toward bridging this divide.”
BackgroundTowards Generalist Biomedical AI
Tu, T., Azizi, S., Driess, D., Schaekermann, M., Amin, M., Chang, P.-C., Carroll, A., Lau, C., Tanno, R., Ktena, I., et al. (2024) · NEJM AI · doi:10.1056/aioa2300138 · Cited by 483
Bhayana, R., Krishna, S., & Bleakney, R. R. (2023) · Radiology · doi:10.1148/radiol.230582 · Cited by 460
The current and future state of AI interpretation of medical images
Rajpurkar, P., & Lungren, M. P. (2023) · New England Journal of Medicine · doi:10.1056/nejmra2301725 · Cited by 443 · cited 1×
“2 There is burgeoning interest in CDSS for enhanced imaging 3 , often termed radiomics, in various disciplines such as breast cancer detection 4 , covid detection 5 , diagnosing congenital cataracts 6 , and hidden fracture location 7 .”
BackgroundThe Role of Explanations on Trust and Reliance in Clinical Decision Support Systems
Bussone, A., Stumpf, S., & O'Sullivan, D. (2015) · 2015 International Conference on Healthcare Informatics · doi:10.1109/ichi.2015.26 · Cited by 441
Long, E., Lin, H., Liu, Z., Wu, X., Wang, L., Jiang, J., An, Y., Lin, Z., Li, X., Chen, J., et al. (2017) · Nature Biomedical Engineering · doi:10.1038/s41551-016-0024 · Cited by 358 · cited 1×
“2 There is burgeoning interest in CDSS for enhanced imaging 3 , often termed radiomics, in various disciplines such as breast cancer detection 4 , covid detection 5 , diagnosing congenital cataracts 6 , and hidden fracture location 7 .”
BackgroundStrohm, L., Hehakaya, C., Ranschaert, E. R., Boon, W. P. C., & Moors, E. H. M. (2020) · European Radiology · doi:10.1007/s00330-020-06946-y · Cited by 329
Comparing ChatGPT and GPT-4 performance in USMLE soft skill assessments
Brin, D., Sorin, V., Vaid, A., Soroush, A., Glicksberg, B. S., Charney, A. W., Nadkarni, G., & Klang, E. (2023) · Scientific Reports · doi:10.1038/s41598-023-43436-9 · Cited by 319 · cited 1×
“36 GPT-4 was recently introduced as the upgraded model for ChatGPT.”
MethodLiberati, E. G., Ruggiero, F., Galuppo, L., Gorli, M., González-Lorenzo, M., Maraldi, M., Ruggieri, P., Polo Friz, H., Scaratti, G., Kwag, K. H., et al. (2017) · Implementation Science · doi:10.1186/s13012-017-0644-2 · Cited by 301
Thirunavukarasu, A. J., Hassan, R., Mahmood, S., Sanghera, R., Barzangi, K., El Mukashfi, M., & Shah, S. (2023) · JMIR Medical Education · doi:10.2196/46599 · Cited by 187
Decker, H., Trang, K., Ramirez, J., Colley, A., Pierce, L., Coleman, M., Bongiovanni, T., Melton, G. B., & Wick, E. (2023) · JAMA Network Open · doi:10.1001/jamanetworkopen.2023.36997 · Cited by 178
As artificial intelligence goes multimodal, medical applications multiply
Topol, E. J. (2023) · Science · doi:10.1126/science.adk6139 · Cited by 132
Khader, F., Müller-Franzes, G., Wang, T., Han, T., Tayebi Arasteh, S., Haarburger, C., Stegmaier, J., Bressem, K., Kuhl, C., Nebelung, S., et al. (2023) · Radiology · doi:10.1148/radiol.230806 · Cited by 107 · cited 1×
“This limitation, known as the "single-modality," is a common issue among many LLMs. 24,25 Advancements in multimodal LLM promise enhanced capabilities and integration with diverse data sources.”
BackgroundNon-task expert physicians benefit from correct explainable AI advice when reviewing X-rays
Gaube, S., Suresh, H., Raue, M., Lermer, E., Koch, T. K., Hudecek, M. F. C., Ackery, A. D., Grover, S. C., Coughlin, J. F., Frey, D., et al. (2023) · Scientific Reports · doi:10.1038/s41598-023-28633-w · Cited by 92 · cited 1×
“10 However, most previous imaging CDSS offers only highlight areas deemed significant by AI, 11–15 providing limited insight into the explanation of the diagnosis.”
Background
Cited by
Hidden flaws behind expert-level accuracy of multimodal GPT-4 vision in medicine
Jin, Q., Chen, F., Zhou, Y., Xu, Z., Cheung, J. M., Chen, R., Summers, R. M., Rousseau, J. F., Ni, P., Landsman, M. J., et al. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01185-7 · Cited by 146
“Our fi ndings, therefore, align with the previous ones, which show the expert-level performance of GPT-4V in the closed-book setting 15,16 .”
ResultJin, H. K., Lee, H. E., & Kim, E. (2024) · BMC Medical Education · doi:10.1186/s12909-024-05944-8 · Cited by 85
“Among the studies, two were deemed to be at high risk [33, 34], and 18 at low risk [11, 13, 21, 27, 29–31 ,”
BackgroundInfluentialCollins, B. R., Black, E. W., & Rarey, K. E. (2024) · Clinical Anatomy · doi:10.1002/ca.24178 · Cited by 61
Wada, A., Akashi, T., Shih, G., Hagiwara, A., Nishizawa, M., Hayakawa, Y., Kikuta, J., Shimoji, K., Sano, K., Kamagata, K., et al. (2024) · Diagnostics · doi:10.3390/diagnostics14141541 · Cited by 25
SYNFAC-EDIT: Synthetic Imitation Edit Feedback for Factual Alignment in Clinical Summarization
Mishra, P., Yao, Z., Vashisht, P., Ouyang, F., Wang, B., Mody, V. D., & Yu, H. (2024) · arXiv · doi:10.48550/arxiv.2402.13919 · Cited by 7
Self-citation
Metadata
- Authors
- Yang, Z., Yao, Z., Tasmin, M., Vashisht, P., Jang, W. S., Ouyang, F., Wang, B., Berlowitz, D., & Yu, H.
- Venue
- medRxiv
- Publisher
- openRxiv
- Published
- 2023-10-26
- OpenAlex
- W4387973771
- Cited by
- 36
- References
- 25
- Type
- posted content
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.