AI spots diabetes and stomach cancer from just the colour of tongue

Doctors have long examined patients’ tongues for signs such as changes in colour (a thick white coating can indicate an infection, for instance) or texture (a dry, cracked tongue may be linked to Sjogren’s syndrome, an autoimmune condition).

But scientists have developed artificial intelligence (AI) programs that check the tongue’s colour, texture and shape with impressive accuracy for early signs of diabetes and even stomach cancer.

Now a review of more than 20 studies assessing these programs has concluded that these are so accurate at spotting signs of disease that doctors could soon start using them in hospitals to help diagnose patients, the journal Chinese Medicine reported.

In the most striking of these studies (published in 2024 in the journal Technologies), the AI program correctly diagnosed 58 out of 60 patients with diabetes and anaemia just by assessing a picture of their tongue.

These programs look for tiny changes in someone’s tongue, having been ‘trained’ in what to look for using a database of thousands of photos of tongues of sick patients.

Another study found that AI could spot gastric cancer from subtle tongue colour and texture changes that often accompany stomach disease – such as a thicker coating, patchy colour loss and areas of redness linked to inflammation in the digestive tract.

When tested on new patients, the AI distinguished those with gastric cancer from healthy volunteers with accuracy similar to standard diagnostic tests, such as a gastroscopy (where a tube with a camera is inserted through the mouth and into the stomach) or a CT scan, correctly identifying cases around 85 to 90 per cent of the time, reported eClinicalMedicine in 2023.

‘AI learns by identifying statistical patterns in large collections of tongue images paired with [the patient’s] clinical or health-related data,’ explains Professor Dong Xu, a bioinformatics expert at the University of Missouri.

‘It detects visual characteristics that appear more frequently in individuals with specific conditions than in healthy people, including colour distribution, surface texture, moisture, thickness, coating, fissures and swelling.’

The idea of the tongue being a useful indicator of health is not surprising, say experts.

‘The tongue is referred to as the mirror of general health,’ explains Saman Warnakulasuriya, an emeritus professor of oral medicine and experimental pathology at King’s College London.

‘A smooth dorsal [i.e. the top] tongue may indicate anaemia because when there is insufficient iron, vitamin B12, or folate (vitamin B9), it leads to the loss of papillae [bumps on the tongue that contain taste buds],’ he says.

‘These nutrients are essential for the rapid cell turnover in the tongue’s surface. Without them, the papillae disappear, leaving the tongue smooth and shiny.’

Meanwhile, a dry tongue may be an early symptom of diabetes, as this can lead to dehydration and damage to nerves, reducing saliva production.

High blood sugar levels in the mouth can also promote bacterial and fungal overgrowth, leading to a yellowish coating.

A pale or white tongue can signal anaemia (due to a lack of red blood cells) and a thick white coating could be a sign of infection (the immune response causes the papillae to swell, and bacteria and debris become trapped between them, creating a visible white coating).

Read More

Can AI help doctors make better decisions about prostate cancer treatment?

And ‘hairy leukoplakia’ – white, raised patches with a corrugated or ‘hairy’ appearance on the sides of the tongue (which cannot be scraped off) – can be symptom of the Epstein-Barr virus (which can cause glandular fever), adds Professor Warnakulasuriya.

But GPs may only see a few tongue changes in day-to-day practice – while AI systems, trained on thousands of clinical photographs, can recognise patterns that clinicians may not be familiar with (or that could be too small to be seen by the naked eye) and can then suggest taking a closer look.

‘The availability of clinical pictures in a well-trained AI program could give doctors confidence to narrow down a correct diagnosis,’ says Professor Warnakulasuriya.

But while AI can spot visual patterns, it doesn’t understand what causes them and may make mistakes.

For instance, AI might associate a pale tongue with anaemia because many anaemic patients in its training data had pale tongues, but a pale tongue could be caused by other factors such as poor circulation.

Also, an experienced doctor can consider the patient’s full medical history, symptoms, lifestyle factors and other clinical findings to determine whether a tongue abnormality is significant or harmless.

AI might flag something as suspicious when it’s actually normal, or miss something important.

This is why – as Bernhard Kainz, a professor in medical image computing at Imperial College London, suggests – the technology is most reliable as a broad health checker.

Professor Xu agrees, and warns that AI is only as good as the data it is trained with.

‘Variability in how the photos are taken [such as variations in lighting conditions and camera quality, and whether the tongue is wet or dry] can substantially affect measurements of colour and texture,’ he says. ‘Tongue appearance is also influenced by diet, hydration, smoking, medications and infections, all of which may obscure disease-related signals.’

Even with the best AI, experts stress a tongue scan should never be treated as a diagnosis.

‘It is always necessary to confirm the diagnosis by conducting appropriate laboratory tests,’ says Professor Warnakulasuriya.

‘Used appropriately’ AI tongue analysis ‘can help prioritise care and reduce missed early signs, but it should complement, not replace, established diagnostic pathways and clinical judgment’, adds Professor Kainz.