A recent study demonstrated the potential to detect vocal cord lesions through a patient’s voice, opening the door for artificial intelligence applications that can identify early warning signs of laryngeal cancer from voice recordings. Such lesions may be benign or indicate the early stages of laryngeal cancer.
The research was conducted by scientists from the Department of Informatics and Clinical Epidemiology at Oregon Health & Science University in the United States. The findings were published on August 12 in the journal Frontiers in Digital Health and covered by EurekAlert.
A Global Health Burden
Laryngeal cancer represents a significant global health burden, with approximately 1.1 million cases worldwide in 2021 and nearly 100,000 deaths. Risk factors include smoking, alcohol consumption, and human papillomavirus (HPV) infection. The five-year survival rate ranges between 35% and 78%, depending on tumor stage and location within the larynx.
Early detection is critical for patient safety. Currently, laryngeal cancer diagnosis relies on video nasoendoscopy and biopsies, which are invasive, time-consuming, and may require specialist access, often causing delays in diagnosis.
Study and Voice Biomarkers
Dr. Philip Jenkins, postdoctoral researcher in clinical informatics and co-author of the study, stated:
“We show that using a suitable dataset, we can leverage voice biomarkers to distinguish patients with vocal cord lesions from those without.”
Jenkins and his colleagues participated in the Bridge2AI-Voice project, part of the Bridge2AI consortium at the U.S. National Institutes of Health, an initiative applying AI to complex biomedical challenges.
The researchers analyzed variations in pitch, tone, volume, and clarity using the first version of a publicly available voice AI dataset, comprising over 12,000 voice recordings from 306 participants across North America.
A small subset of participants included individuals with known laryngeal cancer, benign vocal cord abnormalities, as well as spasmodic dysphonia and unilateral vocal cord paralysis.
Next Steps
The research team plans to apply the new algorithms to larger datasets and test them in hospitals using patient voices.
Jenkins explained:
“To move from this study to an AI tool capable of detecting vocal cord lesions, we will train models on a larger, expert-labeled dataset and then test the system to ensure it performs equally well for both women and men.”
Currently, clinical trials on voice-based health tools are underway. Jenkins predicts that, with larger datasets and clinical validation, AI tools for early detection of vocal cord lesions could enter pilot testing within the next two years.
