Deep Learning in Otolaryngology: A Narrative Review.

Area of Science:

• Otolaryngology
• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Deep learning (DL) utilizes multilayered neural networks for complex pattern recognition in large datasets.
• DL autonomously learns hierarchical representations, outperforming traditional machine learning in analyzing medical images and physiological signals.
• Clinical integration of DL remains challenging due to its complexity.

Purpose of the Study:

• To review recent applications of DL in otolaryngology.
• To propose a framework for integrating DL into otolaryngology clinical practice.
• To synthesize the potential and challenges of DL in the field.

Main Methods:

• A narrative review of 327 original research studies on DL in otolaryngology published between 2020 and 2025.
• Articles were categorized into detection/diagnosis, prediction/prognostics, image segmentation, and emerging applications.
• Analysis focused on diagnostic performance, prognostic capabilities, and segmentation accuracy.

Main Results:

• DL models demonstrated high accuracy in identifying nasopharyngeal carcinoma (92%), laryngeal neoplasms (86%), and otologic pathology (>95%).
• Applications include survival stratification, recurrence prediction, anatomical region delineation, hearing aid optimization, and surgical instrument tracking.
• Proof-of-concept studies show DL performance comparable to expert clinicians.

Conclusions:

• DL holds significant potential to enhance diagnostic accuracy, predict patient outcomes, and guide intraoperative procedures in otolaryngology.
• Widespread adoption necessitates high-quality, representative datasets, mitigation of algorithmic bias, and robust model interpretability.
• Emerging frameworks like federated learning and explainability can foster clinician trust and ensure DL tools meaningfully contribute to patient care.