AI-Powered Laryngoscopy: Exploring the Future With Google Gemini.

Area of Science:

• Artificial Intelligence in Medicine
• Otolaryngology
• Medical Imaging Analysis

Background:

• Foundation models (FMs) are versatile AI systems trained on diverse data.
• Multimodal FMs can process various data types, including video.
• AI's role in clinical decision support is rapidly expanding.

Purpose of the Study:

• To evaluate Google Gemini 1.5 Pro's performance in interpreting laryngoscopy images and videos.
• To assess the potential of advanced foundation models in otolaryngology.

Main Methods:

• Google Gemini 1.5 Pro was used to analyze laryngoscopy frames and videos from Google Images and YouTube.
• The model's ability to recognize the procedure, diagnose pathology, and identify lesion sides was tested.
• Performance was evaluated without specific fine-tuning for the task.

Main Results:

• Gemini 1.5 Pro recognized laryngoscopy in 98.9% of frames and 100% of videos.
• Accurate pathology diagnosis was achieved in 62.5% of frames and 20.0% of videos.
• The model identified lesion sides in 65.9% of frames and 40.0% of videos, and narrated operative videos.

Conclusions:

• Google Gemini 1.5 Pro demonstrates significant potential for analyzing laryngoscopy procedures.
• Foundation models show promise as clinical decision support tools in specialized medical fields like otolaryngology.