Laryngeal cancer, often referred to as voice box cancer, remains an important global health issue.
In 2021 alone, an estimated 1.1 million people were diagnosed with the disease worldwide, killing around 100,000 lives.
Known risk factors include smoking, heavy alcohol consumption, and infection with human papillomavirus (HPV).
Survival rates vary dramatically, ranging from 35% to 78% over 5 years, depending on the stage of the cancer and its location within the larynx.
Although early diagnosis is important to improve patient outcomes, current detection methods often require invasive procedures and professional referral, causing delays.
Now, innovative research suggests that artificial intelligence (AI) can revolutionize the way in which you can detect illnesses simply by listening to the voice.
Issues for early detection
Currently, the diagnosis of laryngeal cancer includes video nasal endoscopy and tissue biopsy.
Although effective, these methods are uncomfortable, time-consuming and rely on access to trained professionals. This means that a valuable week or months will pass before treatment begins.
However, whether caused by benign nodules, polyps, or early stage cancer, changes in the folding of the voice often change the way a person hears. Detecting these subtle changes could potentially provide at-risk patients with a faster, less invasive warning system.
Turn your voice into a diagnostic tool
Researchers at the Oregon University of Health are working at the National Institutes of Health Bridge 2AI-voice project, demonstrating a proof-of-concept system that uses AI to detect voice folding abnormalities from recorded audio.
The team analyzed 12,523 audio samples from 306 North American participants sourced from the first Public Bridge 2AI-voice dataset.
Some of these participants identified laryngeal cancer, others had benign vocal folding lesions or different speech dysfunction, and no speech status was diagnosed in the control group.
By studying variations in tone, pitch, loudness, and vocal clarity, the researchers have identified acoustic markers that clearly differ significantly between healthy male voices, those with benign lesions, and voices of laryngeal cancer, with distinct harmonic and noise ratios and fundamental frequencies.
Although this first study did not detect similar patterns in women, researchers believe that a larger dataset could reveal comparable markers.
Why harmonic-noise ratio is important
One of the most promising indicators was the fluctuations in the harmonic-noise ratio. This is a measure of how the audio signal is pure tone and random noise.
Lower ratios often show vocal folding irregularities, which constitute a potential early warning sign of laryngeal cancer progression.
Monitoring these changes over time allows physicians to track the clinical evolution of the lesions and detect malignant tumors more quickly without the need for immediate invasive testing.
Roads for AI voice screening
Although this study is still in its early stages, its implications are broad.
The team’s next steps include training AI models on larger, ethically sourced datasets, accurately running across genders, and verifying the system in a real clinical setting.
If successful, AI-powered voice analysis could one day become part of a routine health check that offers a non-invasive, low-cost screening method for laryngeal cancer.
Researchers estimate that pilot testing of such tools will begin within the next few years, potentially converting diagnostic speeds and accessibility around the world.
Source link
