Shawn H Sun, Kasha Chen, Samuel Anavim, Michael Phillipi, Leslie Yeh, Kenneth Huynh, Gillean Cortes, Julia Tran, Mark Tran, Vahid Yaghmai, Roozbeh Houshyar
{"title":"Large Language Models with Vision on Diagnostic Radiology Board Exam Style Questions.","authors":"Shawn H Sun, Kasha Chen, Samuel Anavim, Michael Phillipi, Leslie Yeh, Kenneth Huynh, Gillean Cortes, Julia Tran, Mark Tran, Vahid Yaghmai, Roozbeh Houshyar","doi":"10.1016/j.acra.2024.11.028","DOIUrl":null,"url":null,"abstract":"<p><strong>Rationale and objectives: </strong>The expansion of large language models to process images offers new avenues for application in radiology. This study aims to assess the multimodal capabilities of contemporary large language models, which allow analysis of image inputs in addition to textual data, on radiology board-style examination questions with images.</p><p><strong>Materials and methods: </strong>280 questions were retrospectively selected from the AuntMinnie public test bank. The test questions were converted into three formats of prompts; (1) Multimodal, (2) Image-only, and (3) Text-only input. Three models, GPT-4V, Gemini 1.5 Pro, and Claude 3.5 Sonnet, were evaluated using these prompts. The Cochran Q test and pairwise McNemar test were used to compare performances between prompt formats and models.</p><p><strong>Results: </strong>No difference was found for the performance in terms of % correct answers between the text, image, and multimodal prompt formats for GPT-4V (54%, 52%, and 57%, respectively; p = .31) and Gemini 1.5 Pro (53%, 54%, and 57%, respectively; p = .53). For Claude 3.5 Sonnet, the image input (48%) significantly underperformed compared to the text input (63%, p < .001) and the multimodal input (66%, p < .001), but no difference was found between the text and multimodal inputs (p = .29). Claude significantly outperformed GPT and Gemini in the text and multimodal formats (p < .01).</p><p><strong>Conclusion: </strong>Vision-capable large language models cannot effectively use images to increase performance on radiology board-style examination questions. When using textual data alone, Claude 3.5 Sonnet outperforms GPT-4V and Gemini 1.5 Pro, highlighting the advancements in the field and its potential for use in further research.</p>","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":" ","pages":""},"PeriodicalIF":3.8000,"publicationDate":"2024-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Academic Radiology","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1016/j.acra.2024.11.028","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"RADIOLOGY, NUCLEAR MEDICINE & MEDICAL IMAGING","Score":null,"Total":0}
引用次数: 0
Abstract
Rationale and objectives: The expansion of large language models to process images offers new avenues for application in radiology. This study aims to assess the multimodal capabilities of contemporary large language models, which allow analysis of image inputs in addition to textual data, on radiology board-style examination questions with images.
Materials and methods: 280 questions were retrospectively selected from the AuntMinnie public test bank. The test questions were converted into three formats of prompts; (1) Multimodal, (2) Image-only, and (3) Text-only input. Three models, GPT-4V, Gemini 1.5 Pro, and Claude 3.5 Sonnet, were evaluated using these prompts. The Cochran Q test and pairwise McNemar test were used to compare performances between prompt formats and models.
Results: No difference was found for the performance in terms of % correct answers between the text, image, and multimodal prompt formats for GPT-4V (54%, 52%, and 57%, respectively; p = .31) and Gemini 1.5 Pro (53%, 54%, and 57%, respectively; p = .53). For Claude 3.5 Sonnet, the image input (48%) significantly underperformed compared to the text input (63%, p < .001) and the multimodal input (66%, p < .001), but no difference was found between the text and multimodal inputs (p = .29). Claude significantly outperformed GPT and Gemini in the text and multimodal formats (p < .01).
Conclusion: Vision-capable large language models cannot effectively use images to increase performance on radiology board-style examination questions. When using textual data alone, Claude 3.5 Sonnet outperforms GPT-4V and Gemini 1.5 Pro, highlighting the advancements in the field and its potential for use in further research.
期刊介绍:
Academic Radiology publishes original reports of clinical and laboratory investigations in diagnostic imaging, the diagnostic use of radioactive isotopes, computed tomography, positron emission tomography, magnetic resonance imaging, ultrasound, digital subtraction angiography, image-guided interventions and related techniques. It also includes brief technical reports describing original observations, techniques, and instrumental developments; state-of-the-art reports on clinical issues, new technology and other topics of current medical importance; meta-analyses; scientific studies and opinions on radiologic education; and letters to the Editor.