Laryngeal cancer, often called cancer of the voice box, remains a significant global health challenge.

In 2021 alone, an estimated 1.1 million people were diagnosed with the disease worldwide, and around 100,000 lives were lost.

Known risk factors include smoking, heavy alcohol consumption, and infection with human papillomavirus (HPV).

Survival rates vary dramatically, ranging from 35% to 78% over five years depending on the cancer’s stage and its location within the larynx.

Early diagnosis is critical for improving patient outcomes, but current detection methods often require invasive procedures and specialist referrals, causing delays.

Now, an innovative study suggests that artificial intelligence (AI) could revolutionise how we spot the disease – simply by listening to the voice.

The challenge of early detection

At present, diagnosing laryngeal cancer involves video nasal endoscopy and tissue biopsies.

While effective, these methods are uncomfortable, time-consuming, and dependent on access to trained specialists. This can mean valuable weeks or months pass before treatment begins.

However, changes to the vocal folds – whether caused by benign nodules, polyps, or early-stage cancer – often alter the way a person sounds. Detecting these subtle changes could provide an earlier, less invasive warning system for patients at risk.

Turning the voice into a diagnostic tool

Researchers from Oregon Health & Science University, working within the US National Institutes of Health’s Bridge2AI-Voice project, have now demonstrated a proof-of-concept system that uses AI to detect vocal fold abnormalities from recorded speech.

The team analysed 12,523 voice samples from 306 participants across North America, sourced from the first public Bridge2AI-Voice dataset.

Some of these participants had confirmed laryngeal cancer, others had benign vocal fold lesions or different voice disorders, and a control group had no diagnosed voice condition.

By studying variations in tone, pitch, loudness, and vocal clarity, the researchers identified distinct acoustic markers, particularly the harmonic-to-noise ratio and fundamental frequency, that differed significantly between healthy male voices, those with benign lesions, and those with laryngeal cancer.

While similar patterns were not detected in women in this initial study, the researchers believe larger datasets could reveal equivalent markers.

Why harmonic-to-noise ratio matters

One of the most promising indicators was variation in the harmonic-to-noise ratio – a measure of how much of a voice signal is pure tone versus random noise.

Lower ratios often indicate vocal fold irregularities, making this a potential early warning sign for laryngeal cancer progression.

Monitoring these changes over time could enable doctors to track the clinical evolution of lesions and detect malignancies sooner, without the need for immediate invasive testing.

The road ahead for AI voice screening

Although this research is still at an early stage, the implications are far-reaching.

The team’s next steps involve training the AI model with larger, ethically sourced datasets, ensuring it performs accurately across genders, and validating the system in real-world clinical environments.

If successful, AI-powered voice analysis could one day be part of routine health checks, offering a non-invasive, low-cost screening method for laryngeal cancer.

The researchers estimate that pilot testing of such tools could begin within the next few years, potentially transforming the speed and accessibility of diagnosis worldwide.