Deep learning-based allergic rhinitis diagnosis using nasal endoscopy images.

Area of Science:

• Otolaryngology
• Medical Imaging
• Artificial Intelligence

Background:

• Allergic rhinitis (AR) diagnosis traditionally relies on skin prick tests (SPT), which correlate poorly with symptoms and are invasive.
• Current machine learning diagnostic methods use structured clinical data, not visual information from nasal endoscopy.

Purpose of the Study:

• To develop a quantitative, non-invasive diagnostic method for allergic rhinitis using deep learning analysis of nasal endoscopy images.
• To investigate the potential of analyzing inferior turbinate color distribution and features for AR detection.

Main Methods:

• Utilized deep learning to analyze nasal endoscopy images, focusing on the inferior turbinate region.
• Employed CIE-Lab color space and adaptive histogram features for quantitative analysis.
• Extracted features using Convolutional Neural Networks (CNNs) and histograms, followed by classification with Support Vector Machines (SVM) and fully connected classifiers.

Main Results:

• The proposed deep learning model achieved a diagnostic accuracy of 90.80% for images showing allergic rhinitis symptoms.
• Demonstrated the feasibility of using optical analysis of turbinate color for AR diagnosis.

Conclusions:

• A novel approach using deep learning on nasal endoscopy images offers a promising, quantitative method for diagnosing allergic rhinitis.
• Further research with larger datasets and broader image inclusion can enhance model robustness and validate optical analysis as a non-invasive AR diagnostic tool.