Particle.news

Download on the App Store

AI Voice Analysis Distinguishes Laryngeal Cancer and Vocal Fold Lesions in Men

Researchers call for expansion to professionally labeled, demographically varied datasets before voice-based cancer screening can advance to clinical pilots

Image
Image
Image
man with mouth open

Overview

  • A proof-of-principle study published August 12 finds that harmonic-to-noise ratio and pitch metrics separate healthy men, men with benign vocal fold lesions and men with laryngeal cancer based on 12,523 recordings
  • No informative acoustic markers emerged for women in the Bridge2AI-Voice dataset, highlighting a significant gap in current AI screening research
  • Small subgroup sizes for benign lesions (13 cases) and cancers (10 cases) limit the statistical power and generalizability of the findings
  • Authors urge assembly of larger, professionally labeled, multi-institutional voice datasets with balanced demographics to train robust screening algorithms
  • Pending dataset expansion and clinical validation, pilot testing of a non-invasive, AI-assisted voice triage tool could begin within two years