Oregon, United States (Enmaeya News) — Researchers at Oregon Health & Science University have demonstrated that vocal cord lesions can be detected through a patient’s voice, opening the door to a new artificial intelligence (AI) application that could recognize early warning signs of laryngeal cancer from voice recordings. Such lesions may be benign, but they can also indicate early stages of laryngeal cancer.
The study, published Aug. 12 in Frontiers in Digital Health and reported by EurekAlert, underscores the potential for noninvasive early detection of a disease that poses a significant public health burden.
Worldwide, an estimated 1.1 million cases of laryngeal cancer were reported in 2021, resulting in nearly 100,000 deaths. Risk factors include smoking, alcohol use, and infection with human papillomavirus. The five-year survival rate for laryngeal cancer patients ranges from 35% to 78%, depending on tumor stage and location.
Early detection is critical. Currently, laryngeal cancer is diagnosed through video nasoscopy and biopsies, which are invasive procedures requiring specialist care. Limited access to qualified clinicians can delay diagnosis, reducing patient outcomes.
Dr. Philip Jenkins, a postdoctoral researcher in clinical informatics at Oregon Health & Science University and co-author of the study, said, “We demonstrate here that using a dataset, we can utilize voice biomarkers to distinguish the voices of patients with vocal cord lesions from those without.”
Jenkins and colleagues are part of the “Bridge2AI-Voice” project within the Bridge2AI consortium, a U.S. National Institutes of Health initiative aimed at applying AI to complex biomedical challenges.
For the study, researchers analyzed variations in pitch, tone, volume, and clarity using the first version of a public AI voice dataset, which included over 12,000 voice recordings from 306 participants across North America. Among the participants, a small number had known laryngeal cancer, benign vocal cord lesions, or other conditions, including spasmodic dysphonia and unilateral vocal cord paralysis.
The researchers plan to expand their work by applying new AI algorithms to larger datasets and testing them on patients’ voices in hospital settings. Jenkins said, “To move from this study to an AI tool that can detect vocal cord lesions, we will train models using a larger set of voice recordings, classified by specialists, and then test the system to ensure it works equally well for women and men.”
Currently, trials of voice-based health tools are underway. Jenkins predicts that with larger datasets and clinical validation, similar AI tools for detecting vocal cord lesions could enter pilot testing within the next two years.
