在CT肺血管图像上对肺栓塞识别的两个领先的大型语言模型进行基准测试
Nitin Chetla1, Tamer Hage2, Swapna Vaja3
1Allopathic Medicine, University of Virginia School of Medicine, Charlottesville, USA.
Cureus
|October 21, 2025
概括
像GPT-4 Omni和Gemini 2.0这样的大型语言模型 (LLM) 在CTPA扫描上显示在检测肺栓塞 (PE) 上有偏差的性能. GPT-4o在识别PE方面表现出色,但在负面情况下扎,而Gemini 2.0在排除PE方面表现更好.
科学领域:
- 人工智能在医学中的应用
- 医学成像分析 医学成像分析
- 放射学 放射学是一门学科.
背景情况:
- 大型语言模型 (LLM) 越来越能够进行医学图像解释.
- 在计算机断层扫描肺血管造影 (CTPA) 上检测肺栓塞 (PE) 是一个关键的诊断任务.
- 评估放射学LLM的表现需要标准化的测试方法.
研究的目的:
- 评估GPT-4 Omni (GPT-4o) 和Gemini 2.0在CTPA图像上检测PE的诊断性能.
- 通过使用USMLE样式提示,比较这些LLM识别PE阳性和PE阴性病例的能力.
- 了解目前LLMs在放射学解释中的局限性和诊断偏差.
主要方法:
- 来自RSNA PE检测挑战赛2020年的CTPA图像被转换为PNG格式.
- GPT-4o和双子座2.0分析了PE阳性和PE阴性CTPA切片的大数据集.
- 图像通过设计用于分类响应的API提示来呈现,模仿USMLE Step 1格式.
主要成果:
- 对于PE检测,GPT-4o具有高灵敏度 (81%),但特异性低 (10%).
- 双子座2.0在PE检测方面表现出高特异性 (98%),但低灵敏性 (14%).
- F1评分表明GPT-4o在识别阳性病例中的强度 (0.59) 和双子座2.0在消极病例中的强度 (0.70).
结论:
- GPT-4o和双子座2.0表现出相反的诊断偏差,影响了它们在PE检测中的实用性.
- 目前,LLM在实现放射性诊断的平衡灵敏度和特异性方面存在局限性.
- 为了安全有效地整合到临床放射学工作流程中,需要进一步完善LLMs.
相关概念视频
01:29Pulmonary Embolism II: Diagnostic Studies and Interprofessional Care
311
Diagnosing Pulmonary EmbolismDiagnosing pulmonary embolism (PE) involves clinical assessment and advanced imaging tests. The preferred diagnostic tool is the spiral (helical) CT scan or CT angiography (CTA), which uses intravenous contrast media to visualize the pulmonary vasculature and identify emboli.A ventilation-perfusion (V/Q) scan is an alternative for patients unable to receive contrast media. This scan includes both perfusion and ventilation scanning. Perfusion scanning involves...
311
01:13Radiological Investigation III: Pulmonary Angiogram and PET Scan
402
Radiological investigations are paramount in the diagnosis and management of various pulmonary diseases. Two essential investigations are the Pulmonary Angiogram and the Positron Emission Tomography (PET) Scan.
Pulmonary Angiogram
A Pulmonary Angiogram is an invasive procedure involving injecting a contrast medium through a catheter threaded into the pulmonary artery or the right side of the heart to visualize the pulmonary vasculature. Computed Tomography (CT) scans have mainly replaced this...
Pulmonary Angiogram
A Pulmonary Angiogram is an invasive procedure involving injecting a contrast medium through a catheter threaded into the pulmonary artery or the right side of the heart to visualize the pulmonary vasculature. Computed Tomography (CT) scans have mainly replaced this...
402
01:29Pulmonary Embolism I: Introduction
504
Pulmonary embolism (PE) occurs when a thrombus, fat or air embolus, amniotic fluid, or tumor tissue blocks one or more pulmonary arteries. These blockages originate in the venous system or the right side of the heart.EtiologyPE primarily arises from deep vein thrombosis (DVT) and other hypercoagulable states, such as inherited thrombophilias. Additional etiological factors include venous stasis, commonly seen in obesity, and endothelial injury from surgery and trauma. Less common causes include...
504
01:28Imaging Studies for Cardiovascular System V: CT
279
Cardiac computed tomography (CT) scanning is an advanced cardiac imaging technique that utilizes CT technology, with or without intravenous (IV) contrast, to produce accurate cross-sectional virtual slices of specific areas of the heart, coronary circulation, and major blood vessels such as the aorta, pulmonary veins, and arteries. The computer processes these slices to generate three-dimensional images. Multidetector CT (MDCT) is a rapid form of CT scanning that captures multiple slices...
279