High-Speed Videoendoscopy and Stiffness Mapping for AI-Assisted Glottic Lesion Differentiation.

Area of Science:

• Otolaryngology
• Medical Imaging
• Computational Biology

Background:

• Accurate differentiation between benign and malignant glottic lesions is crucial for effective treatment.
• Current diagnostic methods may be invasive or lack objective assessment of vocal fold biomechanics.

Purpose of the Study:

• To evaluate high-speed videoendoscopy (HSV) for non-invasively differentiating benign and malignant glottic lesions.
• To introduce and assess a new parameter, vocal fold stiffness (SAI), for objective vocal fold biomechanical analysis.
• To explore the utility of machine learning models in improving diagnostic accuracy.

Main Methods:

• High-speed videoendoscopy (HSV) was performed on 102 participants (21 controls, 39 benign lesions, 42 glottic cancer).
• Quantified laryngotopographic (SAI) and kymographic parameters related to vocal fold stiffness, amplitude, symmetry, and glottal dynamics.
• Utilized receiver operating characteristic (ROC) analysis and machine learning (SVM classifier) for statistical analysis and lesion classification.

Main Results:

• Univariate ROC analysis showed high efficacy for SAI (AUC=0.91) and amplitude asymmetry (AUC=0.92) in distinguishing normophonic from organic lesions.
• Machine learning models, particularly SVM, achieved improved diagnostic performance with an AUC of 0.93 for detecting organic lesions.
• The SVM classifier demonstrated an AUC of 0.83 for differentiating benign from malignant glottic lesions.

Conclusions:

• The study highlights the potential of the vocal fold stiffness parameter (SAI) derived from HSV as a non-invasive tool.
• SAI can aid in supporting histopathological evaluation for laryngeal lesions.
• Machine learning integration significantly enhances the diagnostic performance of HSV-based parameters for glottic lesions.