Vural Camalan Burcu, Doluoglu Sumeyra, Taraf Nazlim Hilal, Gunay Mehmet Murat, Ozlugedik Samet
Department of Otorhinolaryngology Head and Neck Surgery, University of Health Sciences Ankara Etlik City Hospital, Varlık Mahallesi, Halil Sezai Erkut Caddesi, No:5, 06170, Yenimahalle, Ankara, Turkey.
Eur Arch Otorhinolaryngol. 2025 Jun 17. doi: 10.1007/s00405-025-09524-4.
This prospective simulation study was conducted to evaluate and compare the performance of ChatGPT (o1, 2023) and DeepSeek (V3, 2024) in staging and treatment planning for head and neck cancers.
This prospective simulation study was conducted in March 2025 to evaluate and compare the performance of two advanced artificial intelligence (AI) models, ChatGPT (o1, 2023) and DeepSeek (V3, 2024), in clinical decision-making for head and neck malignancies. A total of 50 hypothetical, guideline-based clinical scenarios were carefully designed in English by two otorhinolaryngologists in alignment with the National Comprehensive Cancer Network® (NCCN®) Guidelines Version 2.2025.
In the overall analysis of treatment planning performance, DeepSeek (V3, 2024) demonstrated statistically superior accuracy compared to ChatGPT (o1, 2023) (p = 0.04). Both models showed comparable performance in tumor staging (p = 0.83). Both DeepSeek (p = 0.0001) and ChatGPT (p = 0.02) were statistically successful in respect of staging accuracy and providing fully correct answers on the subject of treatment.
Although DeepSeek V3 demonstrated promising capability for clinical decision support in head and neck oncology, these artificial intelligence tools cannot replace multidisciplinary tumor boards. However, they can significantly streamline clinical workflows by rapidly organizing patient data, thereby enhancing board efficiency. Future efforts should prioritize the development and integration of secure, institution-specific, local large language models tailored for oncological decision-making.
本前瞻性模拟研究旨在评估和比较ChatGPT（版本1，2023）和DeepSeek（版本3，2024）在头颈部癌症分期和治疗计划中的性能。
本前瞻性模拟研究于2025年3月进行，以评估和比较两种先进的人工智能（AI）模型ChatGPT（版本1，2023）和DeepSeek（版本3，2024）在头颈部恶性肿瘤临床决策中的性能。两名耳鼻喉科医生按照美国国立综合癌症网络（NCCN）2025年第2版指南，精心设计了50个基于指南的假设临床病例，并用英文呈现。
在治疗计划性能的总体分析中，与ChatGPT（版本1，2023）相比，DeepSeek（版本3，2024）在统计学上显示出更高的准确性（p = 0.04）。在肿瘤分期方面，两种模型表现相当（p = 0.83）。在分期准确性和提供关于治疗的完全正确答案方面，DeepSeek（p = 0.0001）和ChatGPT（p = 0.02）在统计学上均取得成功。
尽管DeepSeek V3在头颈部肿瘤学临床决策支持方面显示出有前景的能力，但这些人工智能工具不能取代多学科肿瘤委员会。然而，它们可以通过快速整理患者数据显著简化临床工作流程，从而提高委员会效率。未来的工作应优先开发和整合针对肿瘤学决策的安全、特定机构的本地大语言模型。
