European archives of oto-rhino-laryngology : official journal of the European Federation of Oto-Rhino-Laryngological Societies (EUFOS) : affiliated with the German Society for Oto-Rhino-Laryngology - Head and Neck Surgery · 2026
Turing problems in otolaryngology: a scoping review of the principal challenges of artificial ıntelligence and large language models.
СтатьяОбзорНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Sinanoglu Nidanur, Ismayilova Azada, Muradova Antiga, Hajiyeva Rovsahana, Ahmadov Natig, Hajiyev Yusif, Aliyeva Aynur
Аннотация
BACKGROUND: Artificial intelligence (AI) and large language models (LLMs) have rapidly entered otolaryngology-head and neck surgery (OHNS). Despite accelerating publication output, critical translational and safety challenges remain undercharacterized.
METHODS: A scoping review was conducted, and PubMed/MEDLINE, Cochrane Library, Embase, Web of Science, and Scopus were searched from January 2020 to June 2025 using pre-specified terms encompassing AI, LLMs, machine learning, and deep learning in OHNS.
RESULTS: Of 3,648 screened records, 68 met the final inclusion criteria. Six principal challenge domains were identified: (1) accuracy and validity (LLM correct-answer rates: 53-75% across studies); (2) hallucination and reference fabrication, including a 61.6% reference-to-prompt irrelevancy rate across the platforms evaluated in one study; (3) the 'AI Chasm' translational gap (99.3% of deep-learning studies remained in silico); (4) Black-Box/explainability failure; (5) algorithmic bias and demographic disparities; and (6) data privacy, regulatory compliance, and legal accountability. GPT-4-class models consistently outperformed GPT-3.5, and domain-specific models (e.g., ChatENT) achieved error reductions of 26-58%.
CONCLUSIONS: Current AI and LLM tools in OHNS demonstrate promising but insufficient accuracy for unsupervised clinical deployment. Structured governance frameworks, mandatory clinical validation pipelines, and bias-audited datasets are urgently required.
METHODS: A scoping review was conducted, and PubMed/MEDLINE, Cochrane Library, Embase, Web of Science, and Scopus were searched from January 2020 to June 2025 using pre-specified terms encompassing AI, LLMs, machine learning, and deep learning in OHNS.
RESULTS: Of 3,648 screened records, 68 met the final inclusion criteria. Six principal challenge domains were identified: (1) accuracy and validity (LLM correct-answer rates: 53-75% across studies); (2) hallucination and reference fabrication, including a 61.6% reference-to-prompt irrelevancy rate across the platforms evaluated in one study; (3) the 'AI Chasm' translational gap (99.3% of deep-learning studies remained in silico); (4) Black-Box/explainability failure; (5) algorithmic bias and demographic disparities; and (6) data privacy, regulatory compliance, and legal accountability. GPT-4-class models consistently outperformed GPT-3.5, and domain-specific models (e.g., ChatENT) achieved error reductions of 26-58%.
CONCLUSIONS: Current AI and LLM tools in OHNS demonstrate promising but insufficient accuracy for unsupervised clinical deployment. Structured governance frameworks, mandatory clinical validation pipelines, and bias-audited datasets are urgently required.