Classifying Vocal Folds Fixation from Endoscopic Videos with Machine Learning

Area of Science:

• Otolaryngology
• Medical Imaging
• Machine Learning

Background:

• Vocal fold motility assessment is crucial for diagnosing functional deficits and staging glottic cancer.
• Current diagnostic endoscopy relies on subjective interpretation of videoendoscopic frames.
• A need exists for objective, reliable, and repeatable methods for vocal cord motility evaluation.

Purpose of the Study:

• To propose and evaluate a machine learning (ML) approach for objective classification of vocal cord motility.
• To develop a computer-assisted tool to support clinicians in vocal cord function assessment.

Main Methods:

• A dataset of 558 images was extracted from endoscopic videos of 186 patients with normal or fixed vocal cords.
• Features were extracted from images to train and test four ML classifiers.
• XGBoost classifier was utilized for vocal cord motility classification.

Main Results:

• The XGBoost model achieved high performance metrics: precision = 0.82, recall = 0.82, F1 score = 0.82, and accuracy = 0.82.
• Comparison of ML models indicated XGBoost as the best performing classifier.
• The developed approach demonstrated potential for precise and reliable clinical support.

Conclusions:

• Machine learning offers a promising, objective method for vocal cord motility assessment.
• This computer-assisted approach can enhance the accuracy and reliability of clinical evaluations in otolaryngology.
• The study advances computer-assisted otolaryngology by providing a tool for objective motility assessment.