Diagnostic performance and short-interval reproducibility of multimodal large language models in differentiating cholesteatoma from chronic otitis media using key-image temporal bone high-resolution computed tomography
Diagnostic and Interventional Radiology · 11.08.2026
DOI: 10.4274/dir.2026.264082 ↗Kurz gefasst
Fachartikel in der Zeitschrift Diagnostic and Interventional Radiology, online erschienen am 11.08.2026. Verfasst von Buket Yağcı und 3 weiteren Autor:innen. Beteiligte Einrichtung: University of Health Sciences Türkiye, Antalya Training and Research Hospital, Clinic of Radiology, Antalya, Türkiye.
Abstract (Originalsprache)
PURPOSE: This study aimed to assess whether multimodal large language models (LLMs) can distinguish cholesteatoma from non-cholesteatomatous chronic otitis media (COM) on representative key-image temporal bone high-resolution computed tomography (HRCT) and to evaluate the short-interval reproducibility of their outputs. METHODS: -year radiology resident and a board-certified head and neck radiologist, both unaware of the diagnosis. Subsequently, the same images were analyzed by the GPT-5 and Gemini 2.5 Pro LLMs through their official web interfaces, utilizing structured prompts and a zero-shot approach. These evaluations were conducted in two distinct sessions (S1 and S2) with a 1-week interval. The primary endpoint was accurate binary classification. Accuracy, sensitivity, specificity, positive predictive value, and negative predictive value were calculated with 95% confidence intervals [(CIs); Wilson method] agreement with the reference standard and between sessions and models was assessed with the Cohen kappa (κ) coefficient; and differences in classification were assessed with the McNemar test. RESULTS: = 0.211). CONCLUSION: In this single-center study, GPT-5 and Gemini 2.5 Pro, in the versions evaluated, combined high sensitivity with low specificity and showed only fair-to-moderate between-session reproducibility and exhibited slight inter-model agreement on temporal bone key-image HRCT. These findings do not support their use as independent second readers, and broader generalization to other multimodal LLMs would require the evaluation of additional models. CLINICAL SIGNIFICANCE: The evaluated models lacked the spatial precision and consistency required for the accurate assessment of complex middle ear structures. This finding underscores the necessity for verification by a radiologist and continuous monitoring.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- University of Health Sciences Türkiye, Antalya Training and Research Hospital, Clinic of Radiology, Antalya, Türkiye.
- University of Health Sciences Türkiye, Antalya Training and Research Hospital, Clinic of Otorhinolaryngology, Antalya, Türkiye.
- Akdeniz University Faculty of Medicine, Department of Radiology, Antalya, Türkiye.
ORCID-Profile: Buket Yağcı, Sergen Palaz, Erdem Atalay Çetinkaya, Utku Şenol
Zitieren
APA
Yağcı, B., Palaz, S., Çetinkaya, E. A., & Şenol, U. (2026). Diagnostic performance and short-interval reproducibility of multimodal large language models in differentiating cholesteatoma from chronic otitis media using key-image temporal bone high-resolution computed tomography. Diagnostic and Interventional Radiology. https://doi.org/10.4274/dir.2026.264082
Vancouver
Yağcı B, Palaz S, Çetinkaya EA, Şenol U. Diagnostic performance and short-interval reproducibility of multimodal large language models in differentiating cholesteatoma from chronic otitis media using key-image temporal bone high-resolution computed tomography. Diagnostic and Interventional Radiology. 2026. doi:10.4274/dir.2026.264082
BibTeX
@article{Yac2026,
author = {Yağcı, Buket and Palaz, Sergen and Çetinkaya, Erdem Atalay and Şenol, Utku},
title = {Diagnostic performance and short-interval reproducibility of multimodal large language models in differentiating cholesteatoma from chronic otitis media using key-image temporal bone high-resolution computed tomography},
journal = {Diagnostic and Interventional Radiology},
year = {2026},
doi = {10.4274/dir.2026.264082}
}
EUVIMED