Machine Learning-Enhanced Clinical Decision Support for Diagnosing Sinusitis With Nasal Endoscopy
Dipesh Gyawali1, Thomas Mundy2, Majid Hosseini3
1Department of Otorhinolaryngology, Ochsner Health, New Orleans, Louisiana, USA.
International Forum of Allergy & Rhinology
|October 15, 2025
Summary
A new machine learning framework improves sinusitis diagnosis using nasal endoscopy (NE) by accurately identifying anatomical landmarks and mucus. This AI tool reduces variability and matches otolaryngologist accuracy for better patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Otolaryngology
Background:
- Sinusitis diagnosis relies on nasal endoscopy (NE), but accuracy is hampered by inter-operator variability in identifying anatomical landmarks and mucus.
- Developing objective diagnostic tools is crucial for standardizing sinusitis assessment.
Purpose of the Study:
- To create a novel multi-class machine learning (ML) framework for detecting anatomical structures and mucus in NE images.
- To develop a rule-based clinical algorithm for sinusitis diagnosis using ML-identified features.
Main Methods:
- A YOLOv11-nano model was trained on 3513 NE images annotated for middle turbinate, inferior turbinate, and mucus.
- A rule-based algorithm incorporating anatomical Intersection over Union (IoU) was developed for middle meatus localization and sinusitis classification.
- The system was validated on 178 images from patients with chronic rhinosinusitis without polyps (CRSsNP).
Main Results:
- The ML model achieved over 75% F1 score for detecting turbinates with mucus.
- The clinical algorithm demonstrated 75.0% sensitivity, 76.0% specificity, and 75.2% accuracy for sinusitis classification, with an 81.8% F1 score.
- The framework operated at over 20 frames per second (fps) on a GPU, indicating near real-time performance.
Conclusions:
- The developed ML-driven framework and rule-based algorithm enhance sinusitis diagnosis via NE, reducing inter-operator variability.
- The system's performance rivals that of trained otolaryngologists and offers real-time processing capabilities.
- This technology has the potential to standardize sinusitis care and improve patient outcomes, with future work focusing on diverse sinusitis phenotypes and clinical implementation.
