AI could improve cancer diagnosis – if we get these 5 things right

When it comes to healthcare, we all want to spend more time talking with our doctors and less waiting for test results. Could artificial intelligence (AI) make that the reality? 

There’s been plenty of talk recently about AI and what it could mean for our public services. Importantly for us, research is suggesting it could be used alongside doctors to diagnose more cancers more quickly. With the NHS continuing to miss cancer waiting times targets, and the burden of cancer projected to grow to half a million new cases a year by 2040, that could make all the difference.  

The technology is still developing, but it’s time to start thinking about what we need to do to make AI-assisted cancer diagnosis possible.  

Firstly, what is artificial intelligence? 

AI systems are computer models that can perform tasks traditionally associated with human intelligence. In most cases, they do this by finding patterns in large amounts of inputted data and using this learning to make predictions.  

You’re probably already using AI every day. It’s how your phone recognises your face and directs you from A to B. Those are different skills, and they require different types of AI. Some of the most famous AI models – like ChatGPT – are specifically designed to be creative. They’ve grabbed headlines because, like people, they can surprise you and seem imaginative. The types of AI used in health diagnostics are different. They’re not designed to be creative; their job is to be very accurate. 

The NHS is already using some of them – from heart rhythm monitors and urine sample analysers to smart dictation devices and stroke recognition tools. 

And how could AI help with cancer diagnosis? 

Cancer diagnosis is complex. Specialist doctors such as radiologists and pathologists are trained over many years to detect abnormalities in a scan or a microscope image which could indicate cancer or other problems.  

To make an AI model, developers give it a high-speed version of this training. An AI designed to identify breast cancer from mammograms will have been trained on millions of previous mammograms – each labelled as to whether they are healthy or showing signs of disease. Through this process, it will start to recognise patterns in the images, learning what looks healthy and what could be a sign of cancer. 

When all that training’s done, what does it mean for patients?  

The experience of having a scan or biopsy shouldn’t change at all. But the image or sample will be looked at by an AI system alongside a human doctor. They can work together in a couple of different ways. The AI could instantly highlight parts of tests that it identifies as worrying so a human can spot them more quickly, or it could help offer a fast second opinion if a doctor wants a further check.  

Why would we want to use AI? 

The cancer diagnostic workforce is under intense pressure to keep up with patient need.  

There is currently a shortage of both radiologists and pathologists. Modelling suggests that without changes, these shortfalls will get worse over the next 15 years. 

This problem is made even worse by the fact that demand for diagnostic services is forecast to grow. Cancer Research UK’s modelling suggests that the average number of new cases diagnosed each year is projected to reach half a million by 2040, up from around 385,000 between 2017 and 2019.  

If there are too many cancer tests to analyse, and not enough specialists to analyse them, then there’s a risk of delaying diagnosis. That can have serious consequences for patients. The earlier a cancer is diagnosed, the more likely it is that doctors can treat it successfully. On average, treatments given earlier should also be cheaper for the NHS.  

The NHS is already missing cancer waiting time targets. Without action, there’s very little chance that it will start meeting them. If AI can perform some diagnostic tasks, it could help ease workforce pressures and reduce backlogs. And crucially for patients, it could free up doctors’ time. This would let them do things only they can do, like looking at complex cases, spending more time talking with patients and improving the general experience of care.  

What does the NHS need to do to use AI for cancer diagnosis?

Although AI technology isn’t ready to be used everywhere today, it’s improving fast. Bringing it into cancer diagnosis could be the next big step. Here are the key things the NHS, the UK Government and AI developers need to do to prepare. 

1. Support the workforce 

Today, the NHS doesn’t have all the technical knowhow it needs to effectively introduce AI to cancer diagnosis. These tools are complicated: it will take full-time specialists to set them up, maintain them and troubleshoot them. 

Recent projections suggest that, based on current trends, the NHS will be short of about 17,800 digital and data specialists by 2030. Due to limitations in the current NHS Electronic Staff Record, there’s no way to accurately identify who works in digital roles. We need this information for effective workforce planning. 

The UK Government’s upcoming national digital workforce strategy needs lay out exactly how these missing technicians will be trained and recruited. Employers everywhere are looking to hire people with digital skills, so the NHS needs to make its digital jobs more attractive, with clear career pathways, to compete. 

Current NHS staff also need to feel confident using AI tools in their day-to-day work. That will take training. Estimates suggest 90% of NHS workers will need digital skills over the next 20 years. However, at present 43% of NHS staff feel that they can’t access the right learning and development opportunities when they need to, on any topic. 

NHS employers need to provide training opportunities to make sure staff are comfortable with AI tools and understand how to use them appropriately. 

2. Build the right infrastructure  

AI systems need digital data. Radiology in the UK is already fully digitised, which means it’s in a good position to embrace AI. Most pathologists, however, still look at samples directly under a microscope. In digital pathology, the glass slides are scanned and then viewed on a computer instead. Before AI can assist pathologists in diagnosing cancer from biopsy samples, the NHS needs to invest in a more extensive digital pathology system.  

That’s not all. AI tools often require data from multiple IT systems, like imaging databases or electronic patient records. We want these IT systems to be able to communicate with each other, so they can automatically access the data they need to do their job, rather than relying on people to transfer it across. Unless they can do this – be ‘interoperable’ – then they won’t save much time. NHS trusts need to make sure they’re using the up-to-date IT infrastructure AI systems need to work efficiently.

Putting all the data an AI might need in one place is another way to help with this. Across the entirety of Wales, radiology images and pathology test results are already accessible through one central portal. By joining services up like this, the NHS can help with both the development and deployment of AI. It would also be good news for those affected by cancer, as it would make sharing test results easier. 

3. Nurture confident leaders  

AI might be smart, but it won’t make its way into the health system without a bit of guidance. NHS leaders must see the potential of AI tools and, once they’re ready, choose to use them. That will take bravery, as well as money and manpower. At present, though, AI is too often seen as a ‘nice-to-have’ rather than as part of the solution to large strategic problems. 

Part of the issue is that the leaders of NHS Trusts are overworked and need additional resources, support and incentive to implement innovations. Some simply may not have enough knowledge of, or exposure to, cutting-edge specialist topics such as AI.  

The NHS could address that by training digital leaders who are confident in how to prioritise and most effectively utilise AI solutions for issues such as cancer diagnostics.  

And for those leaders who need assistance now, the NHS, the Department of Health and Social Care, and government agencies should provide clearer guidance about which AI uses it wants to prioritise. They could also help by highlighting AI tools that have already been approved for use and clearly evaluated. That’s worked well in stroke networks. By providing a clear framework for how clinicians can use AI to diagnose strokes, the NHS has helped make sure the new technology benefits patients. 

4. Avoid inequality 

It is crucial that AI helps everyone at risk of cancer, not just certain sections of the population. AI tools need to be able to detect cancer effectively regardless of an individual’s ethnicity, sex, or other characteristics. Otherwise, there is a real risk of exacerbating existing cancer inequalities.  

An AI is only as good as the dataset it was trained on. For example, if an AI that’s meant to detect skin cancers is only trained on examples of fairer skin, it may perform badly at identifying cancers on darker skin.  

AI developers can control for these sorts of unintentional biases by making sure their tools are trained on large, representative datasets. Those in charge of regulating health AI systems also need to remain vigilant to make sure AI are rigorously tested on their ability to avoid biases.  

As diagnostic AI tools become more common, we should also make sure that they are being trained to detect all sorts of cancers – including ones that may disproportionately affect people in more deprived or minority groups. 

Finally, it’s also important that regional inequity doesn’t create a postcode lottery. Because AI is a relatively new technology, there’s a risk that it might only be integrated into research-active hospitals and centres of excellence that already have the necessary expertise, resources and infrastructure. Securing regulatory approval for an innovation does not guarantee uptake in the NHS. In practice, decisions to install and use new technologies are often left to individual trusts or regions. This can lead to disparities in access. The government needs to provide enough money to ensure no areas are left behind when it comes to AI.  

5. Maintain public trust  

AI tools have the potential to improve the lives of cancer patients as well as the clinicians who use them. Despite this, they’ll quickly be abandoned if the public don’t feel they can trust the technology.  

To help people feel comfortable with AI, NHS leaders need to clearly communicate its limitations and potential risks, as well as what we can gain from using it. They also need to listen and respond to the public’s opinions and concerns. It’s especially important that people know how their anonymised health data might be used to train better AI and are free to opt out if they want to.