Avidan Yuval, Sliman Hussein, Weizman Baruch, Ben Court Orel, Fuks Alexander, Adawi Salim, Aker Amir
Department of Cardiology, Lady Davis Carmel Medical Center, 7 Michal St., Haifa, Israel.
Department of Internal Medicine, Lady Davis Carmel Medical Center, Haifa, Israel.
J Ultrasound. 2025 Aug 25. doi: 10.1007/s40477-025-01071-2.
Transesophageal echocardiography (TOE) necessitates meticulous patient preparation. Artificial intelligence (AI) chatbots, such as ChatGPT, hold promise in delivering medical information. This study sought to assess the suitability of the latest iteration, ChatGPT-4 Omni (GPT-4o), as a supplemental resource for providing information on topics related to TOE.
Fifty questions pertaining to patient education, covering both fundamental knowledge (25) and management aspects (25) of TOE results, were submitted to GPT-4o in three separate sessions. Three senior echocardiographers systematically evaluated the responses using a four-point scale, focusing on their appropriateness, clarity, and consistency. Additionally, they assessed confabulations and the omission of relevant content on a binary scale.
Overall, 94% of GPT-4o's responses were rated as "highly appropriate" across 50 tasks, with 98% deemed "highly comprehensible". Responses related to basic knowledge were 96% appropriate and 100% comprehensible, while management-related responses were 92% appropriate and 96% comprehensible. In total, two responses (4%) were rated negatively as quite inappropriate. Minor inconsistencies were observed in 12% of regenerated responses, with relevant content missing in 6% of inquiries.
GPT-4o demonstrated high accuracy and reliability in addressing patient inquiries related to TOE. These findings suggest that this AI platform could effectively supplement patient education and complement healthcare professionals' guidance, thereby reducing the workload of echocardiography lab staff.
经食管超声心动图检查（TOE）需要对患者进行细致的准备。人工智能（AI）聊天机器人，如ChatGPT，有望提供医学信息。本研究旨在评估最新版本的ChatGPT-4 Omni（GPT-4o）作为提供与TOE相关主题信息的补充资源的适用性。
在三个不同的环节中，向GPT-4o提交了50个与患者教育相关的问题，涵盖TOE结果的基础知识（25个）和管理方面（25个）。三位资深超声心动图专家使用四点量表系统地评估回答，重点关注其适当性、清晰度和一致性。此外，他们还以二元量表评估虚构内容和相关内容的遗漏情况。
总体而言，在50项任务中，GPT-4o的回答有94%被评为“非常合适”，98%被认为“非常易懂”。与基础知识相关的回答96%合适，100%易懂，而与管理相关的回答92%合适，96%易懂。总共有两个回答（4%）被负面评为非常不合适。在12%的重新生成的回答中观察到轻微不一致，6%的询问中缺少相关内容。
GPT-4o在回答与TOE相关的患者询问时表现出高准确性和可靠性。这些发现表明，这个人工智能平台可以有效地补充患者教育，并补充医疗保健专业人员的指导，从而减轻超声心动图实验室工作人员的工作量。
