Skip to main content

Together we are beating cancer

Donate now
  • Science & Technology

Can AI help spot early signs of lung cancer in the blood?

Henry Scowcroft
by Henry Scowcroft | In depth

17 November 2026

0 comments 0 comments

Lung cancer cells seen through a microscope, in shades of purple and green,
Lung cancer cells. Photo by: Anne Weston, Francis Crick Institute.

Lung screening with CT scans is transforming outcomes for patients, but it’s not perfect. At the Cancer Research UK National Biomarker Centre in Manchester, Professor Caroline Dive CBE and her team are developing a blood test that could augment lung screening, taking it to the next level.

In a blue-and-white van outside Asda last June, 68-year-old Eddie, a retired fishmonger from Fallowfield, near Manchester, received life-changing news.

The van belonged to the NHS – and it contained a portable scanner. Eddie and his wife Jeanette had arrived for a lung screening appointment, after being invited their GP.

“I very nearly didn’t go,” recalls Eddie, who felt fit and well. But Jeanette insisted: it was no more difficult than a quick trip to the shops.

“Because I smoked as a teenager and had some family history of cancer, they said that I should have a scan there and then,” he remembers.

To the couple’s shock, it found a dark shadow on Eddie’s right lung. A week later, in hospital, he had further tests, including a biopsy – and the couple’s worst fears were confirmed: it was lung cancer. “I can’t even properly remember the conversation. I was in total shock,” Eddie says.

Eddie
Eddie was diagnosed with early stage lung cancer after a lung screening appointment.

But thankfully, Eddie’s tumour had been found early – it was removed with keyhole surgery, and he’ll have regular scans over the next five years to check that it hasn’t returned. His long-term outlook is good – according to one US study, more than 80% of people diagnosed with stage 1 lung cancer through screening were cured by surgery alone. Since its introduction in 2019, NHS England’s screening has scanned more than 800,000 current and former smokers, detecting around 10,700 lung cancers, over three-quarters of which were, like Eddie’s, early-stage, treatable tumours. Before screening, just a third of lung cancers were diagnosed at stage 1 or 2.

But screening has its downsides. False positives can cause stress and anxiety, while regular monitoring after surgery can be a stressful and time-consuming process. There are also potential harms from the radiation used in CT scans, even though the dose is low.

For these reasons, the NHS programme currently only invites current and former smokers, whose higher cancer risk means the benefits outweigh these harms. But estimates suggest as many as 6,000 ‘never-smokers’ – currently ineligible for screening – die from lung cancer every year.

So researchers are looking for ways to make screening more accurate, tell fast-growing tumours apart from slower-growing ones (to spare some people intensive monitoring while offering more intensive therapies to others) and extend its benefits to more people.

At the Cancer Research UK National Biomarker Centre, just minutes from the supermarket where Eddie’s story began, centre director Professor Caroline Dive CBE and her teams are exploring an approach that could transform lung screening.

Using cutting-edge artificial intelligence (AI), they’re looking for answers in people’s blood…

Fishing for clues

“Scans are good at finding people with stage 1 disease, almost all of whom will have surgery and a good chance of cure,” says Dive, who has dedicated much of her career to early cancer detection. “But about one in five will go on and relapse. And that means, in some folks, that an early tumour has already sent out the seeds of metastasis” and begun to spread.

Dive is a world-renowned expert in blood-based cancer biomarkers: subtle changes in the levels or structure of molecules in the blood that reveal the disease’s presence. And at Cancer Research UK, we’re now supporting her team to tackle two key questions in lung screening.

“First, can we use blood tests to find early lung cancers?” says Dive. Early tumours release molecules and cells into the bloodstream – could these form the basis for a blood test, or ‘liquid biopsy’, to augment CT scans and boost their accuracy?

Professor Caroline Dive

But crucially, Dive is also investigating whether this could pick out the faster-growing tumours – something hard to do on a scan. “Can we find a way to tell which of those tumours are ‘born to be bad’, and recur after surgery?” she says, and so help doctors tailor treatment and follow-up monitoring to each patient’s needs.

To investigate, Dive’s team have built a unique, precious resource: thousands of carefully frozen blood samples donated by people during their lung screening appointments in Manchester.

“Since 2016, we’ve been working closely with one of the pioneers of community-based lung screening in Manchester: Professor Phil Crosbie and his team at Wythenshawe Hospital,” says Dive. Together, they hope that, by using sophisticated techniques to compare samples from people with and without early lung cancer, they can identify molecular signatures that could lead to a sensitive and informative blood test.

But with the multitude of different molecules and cells circulating in the blood, knowing which ones signal an early lung cancer is a formidable challenge. To help spot patterns hiding in this this complexity, Caroline’s team are turning to one of the most promising developments in modern research: AI.

Nature’s most complex mixture

Blood is one of nature’s most complex liquids. Billions of cells – chiefly oxygen-carrying red blood cells and various flavours of immune cell – float in a watery soup containing countless biomolecules. These include tens of thousands of different types of proteins; sugars, fats and other metabolites; and fragments of genetic material (DNA and RNA) shed by dying cells –  including cancer cells. And thanks to modern technologies, scientists can now measure the levels and nature of these components with astonishing accuracy.

A scientist's hand as they handle samples. They are wearing a purple glove.

Because these molecules reflect so many of the body’s processes, a well-designed blood test can offer a window into its inner workings, including when we’re ill. But developing a reliable test is challenging. Levels of different molecules naturally fluctuate: with age, infections, allergies, illness, exercise, and even with what we’ve eaten.

The trick, therefore, is separating the signal from the noise.

To do this, Dive’s team have selected 200 pairs of blood samples from their larger collection of screening samples.

“For every sample where screening found an early lung cancer, we’ve been able to match a ‘partner’ sample from someone who remained cancer-free, but who was otherwise similar in age, sex and smoking history,” she explains. This should make the cancer-related differences between samples much easier to distinguish. “We’re fishing in a pond, rather than an entire ocean,” she says.

And rather than relying on one type of measurement – which, Dive says, is “unlikely to get the job done” –  they’re taking a ‘multi-modal’ approach, looking for patterns across a whole range of biological information sources.

For instance, using a technique called proteomics, they’re precisely measuring levels of thousands of circulating proteins. For circulating DNA fragments shed by tumours, they’re looking at three different features: cancer-linked mutations; chemical modifications to the DNA, (including so-called methylation); and differences in the fragments’ size (known as ‘fragmentomics’). They’re looking for rare abnormal circulating cells – including cancer cells –  that might offer clues. And with collaborators in the US, they’re analysing circulating RNA (a shorter-lived cousin of DNA).

Sifting through the data with AI

This will create an enormous dataset, containing hundreds of thousands of measurements from each sample. Finding meaningful, cancer-linked patterns within it is an extraordinary challenge – and that’s where AI comes in. “It’s such a big puzzle, and AI will hugely speed up the process,” says Dive.

The team will first train AI models using data from the first 100 pairs of cancer-positive and matched-negative samples. Then they’ll test whether the best-performing AI model can identify cancer in the second 100 pairs, without the AI ‘knowing’ which samples came from people with cancer. A key challenge will be identifying the minimum number of measurements needed to maintain accuracy across such large datasets.

Then, if successful, the next step will be a much larger study. “If – and it’s a big ‘if’ – the data are convincing, then we’ll need to work with commercial partners and organisations like Cancer Research Horizons,” says Dive. “Developing a biomarker test for routine use is really challenging – and expensive.”

None of this would be possible without the research networks and infrastructure that have developed in Manchester since Dive moved there in 2003, she says. “I’m hugely grateful that Cancer Research UK have given me a stable base to develop strong networks, not only in Manchester with the Christie Hospital and wider NHS trusts, but around the UK and abroad,” she says. “We’ve got the right ingredients in the pot –  now we’re just going to have to see what these precious samples can tell us.”

Meanwhile, Eddie’s most recent scan was clear, and he’s looking to the future. “I had a lot to lose –  I am so glad that my cancer was caught early. Not everyone is that lucky, so research to help others be diagnosed as early, like I was, is so important,” he says. “We must support this crucial work for future generations, so more people can survive their cancer.”

Tell us what you think

Leave a Reply

Your email address will not be published. Required fields are marked *

Read our comment policy.

Tell us what you think

Leave a Reply

Your email address will not be published. Required fields are marked *

Read our comment policy.