Artificial Intelligence Identifies Early Signs of Cancer in Voice Prior to Medical Diagnosis by Doctors
In a significant breakthrough, researchers led by Dr. Phillip Jenkins at Oregon Health & Science University have demonstrated that Artificial Intelligence (AI) can detect early signs of laryngeal cancer through voice recordings. The study, published in the journal Frontiers in Digital Health, utilizes the publicly available Bridge2AI-Voice dataset to identify subtle acoustic changes caused by vocal fold lesions.
The researchers analyzed the acoustic parameters of the voice, including mean fundamental frequency, jitter, variation in pitch, shimmer, and harmonic-to-noise ratio. The study found marked differences in the harmonic-to-noise ratio and fundamental frequency between men without any voice disorder, men with benign vocal fold lesions, and men with laryngeal cancer.
The study included participants with known laryngeal cancer, benign vocal fold lesions, spasmodic dysphonia, and unilateral vocal fold paralysis. It demonstrated the ability to use vocal biomarkers to distinguish voices from patients with vocal fold lesions from those without such lesions.
However, the current results are based on initial exploratory analyses of the Bridge2AI-Voice dataset. The dataset, though publicly available and structured for machine learning, still appears limited in size, especially for female participants. The study found statistically significant acoustic feature differences mostly in cisgender men, with limited statistical evidence in cisgender women due to smaller sample representation.
Clinical validation in larger and more diverse cohorts, including both men and women, is still underway and needed to confirm and generalize these findings across populations. The potential for voice recordings as a biomarker lies in its non-invasive, accessible, and cost-effective nature, potentially allowing early detection before symptoms develop or invasive procedures like endoscopy are needed.
Early diagnosis is crucial, as survival rates for stage 1 laryngeal cancer approach 90%, compared to about 35% in late-stage disease. If refined and validated, a voice-based tool could make early screening for laryngeal cancer vastly easier, potentially catching it after a few seconds of speaking rather than months of hoarseness.
The idea of detecting cancer by voice is part of a growing movement in medicine, where biomarkers are found in everyday signals. The 'Bridge2AI-Voice' project is a nationwide endeavor within the US National Institute of Health's 'Bridge to Artificial Intelligence' consortium. Researchers believe that ethically sourced, large, multi-institutional datasets like Bridge2AI-Voice could help make voice a practical biomarker for cancer risk in clinical care.
To move from the study to an AI tool that recognizes vocal fold lesions, a larger dataset of voice recordings, labeled by professionals, is needed for training models. The system needs to be tested to ensure it works equally well for women and men. The study's corresponding author, Dr. Jenkins, emphasizes the importance of this research, stating, "This approach promises a non-invasive, scalable alternative to current invasive diagnostic methods, improving early detection and patient outcomes."
- Artificial Intelligence (AI) in science and technology has shown potential in detecting early signs of laryngeal cancer through voice recordings, a finding published in Frontiers in Digital Health.
- The study, led by Dr. Phillip Jenkins, identified significant acoustic feature differences for cisgender men with laryngeal cancer, benign vocal fold lesions, and no voice disorder.
- To make voice a practical biomarker for cancer risk in clinical care, larger and more diverse datasets, including those for women, are needed to refine and validate the AI tool.
- Early diagnosis of laryngeal cancer is crucial as survival rates for stage 1 are approximately 90%, compared to about 35% for late-stage disease.
- The 'Bridge2AI-Voice' project, part of the US National Institute of Health's 'Bridge to Artificial Intelligence' consortium, aims to use ethically sourced, large, multi-institutional datasets to help AI tools recognize vocal fold lesions, promising a non-invasive, scalable alternative for early detection.
