A recent study has found that abnormalities of the vocal folds caused by the early stages of laryngeal cancer can alter the acoustics of men’s voices.
The researchers hope AI models could one day be used to help detect laryngeal cancers early on, potentially reducing the time it takes to receive a diagnosis and increasing patients’ survival rate.
“Voice-based health tools are already being piloted,” says author of the study Dr Phillip Jenkins, a postdoctoral fellow at Oregon Health and Science University in the US.
“Here we show that with this dataset we could use vocal biomarkers to distinguish voices from patients with vocal fold lesions from those without such lesions.”
While vocal fold lesions can sometimes be benign polyps, they can also represent the early stages of laryngeal cancer.
Laryngeal cancer, also known as Voice Box cancer, is a type of throat cancer associated with a lump in the neck and difficulties with swallowing and breathing.
The American Cancer Society estimates there will be about 13,000 new cases of laryngeal cancer and approximately 3,910 people will die from the cancer in the US in 2025.
Almost 90% of people survive laryngeal cancer for 5 years if it is detected early at Stage 1. But, if the cancer is detected much later at Stage 4, this figure drops to almost 35% according to Cancer Research UK.
Jenkins and his colleagues used the Bridge2AI-Voice dataset. The collection of 12,523 voice recordings of 306 patients is part of the ‘Bridge to Artificial Intelligence’ consortium, the US National Institute of Health’s proposal to use AI to assist with complex biomedical challenges.
From this data set, the research team analysed 180 recordings from 176 participants. Of this sample, there were 8 women and 8 men who had laryngeal cancer (with or without other vocal cord disorders).
The researchers also analysed voice recordings of patients with ‘spasmodic dysphonia’, a disorder causing involuntary spasms of the vocal cords and with ‘unilateral vocal fold paralysis’, a condition where nerve damage prevents a person from being able to open and close their vocal cords correctly. Others had benign vocal folds lesions.
The researchers concentrated their analysis on variations in pitch within speech and measured the relation between harmonic and noise components of speech, among other acoustic features of the voice.
Jenkins and his team discovered differences in the pitch and harmonic-to-noise ratio between men with laryngeal cancer, men with benign vocal fold lesions and men without any voice disorder.
The authors suggest the harmonic-to-noise ratio can be helpful to monitor the progression of vocal fold lesions in a clinical environment, potentially aiding clinicians in detecting laryngeal cancer at an early stage.
However, they did not pick up on any acoustic features among women with laryngeal cancer, so the detection may only work for men at this stage. The authors suggest a larger dataset may solve this problem.
“Our results suggest that ethically sourced, large, multi institutional datasets like Bridge2AI Voice could soon help make our voice a practical biomarker for cancer risk in clinical care,” says Jenkins.
“To move from this study to an AI tool that recognises vocal fold lesions, we would train models using an even larger dataset of voice recordings, labelled by professionals.
We then need to test the system to make sure it works equally well for women and men.
“Building on our findings, I estimate that with larger datasets and clinical validation, similar tools to detect vocal fold lesions might enter pilot testing in the next couple of years.”
