Classifying Phonotrauma Severity from Vocal Fold Images with Soft Ordinal Regression.

Area of Science:

• Medical Imaging
• Computational Pathology
• Speech Science

Background:

• Phonotrauma, or vocal fold tissue damage, severity assessment is subjective and relies on costly clinical judgment.
• Current methods lack reliability and consistency, hindering large-scale phonotrauma research.

Purpose of the Study:

• To develop the first automated method for classifying phonotrauma severity using vocal fold images.
• To address the ordinal nature of severity labels and inherent label uncertainty in clinical assessments.

Main Methods:

• Utilized an ordinal regression framework to handle the ordered severity levels of phonotrauma.
• Proposed a novel soft ordinal regression loss function to incorporate annotator rating distributions (soft labels).
• Applied the method to classify phonotrauma severity from medical images of vocal folds.

Main Results:

• The soft ordinal regression method achieved predictive performance comparable to that of clinical experts.
• The system produced well-calibrated uncertainty estimates for its severity classifications.
• Demonstrated the feasibility of automated phonotrauma severity assessment.

Conclusions:

• Automated phonotrauma severity classification is achievable with high accuracy.
• The proposed method offers a reliable and cost-effective alternative to subjective clinical judgment.
• This tool can facilitate large-scale phonotrauma studies, advancing clinical understanding and patient care.