Quantitative laryngoscopy with computer-aided diagnostic system for laryngeal lesions

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Otolaryngology

Background:

• Laryngeal lesion diagnosis relies on subjective physician experience.
• Objective criteria are needed for accurate and consistent diagnosis.
• Existing methods lack automated objective analysis.

Purpose of the Study:

• To develop a computer-aided diagnostic system for laryngeal lesions.
• To utilize objective criteria for improved diagnostic accuracy.
• To reduce reliance on subjective physician interpretation.

Main Methods:

• Image compensation for consistent brightness and automatic screening of clear laryngoscopic images.
• Automated segmentation of pharynx and larynx using structural features (ACM).
• Hue and geometric analysis of vocal cords, classified using a support vector machine (SVM) decision tree.

Main Results:

• High detection accuracy for vocal cord polyps (93.15%), cysts (95.16%), leukoplakia (100%), tumors (96.42%), and healthy cords (100%).
• Average test accuracy for laryngeal lesions reached 93.33%.
• Cross-validation accuracy demonstrated robust performance across all classes.

Conclusions:

• Hue and geometric features of the larynx are feasible indicators for identifying lesions.
• The developed system effectively assists physicians in diagnosing laryngeal lesions.
• Objective, AI-driven analysis enhances diagnostic consistency and accuracy.