Artificial intelligence and breast cancer: What's next?

NHS Grampian has been at the forefront of trialling AI in breast cancer screening through its three-year Gemini project, which has led to an "extraordinary" 12% uplift in detection rates from routine mammograms - including finding tiny traces of disease that evaded the human experts.

As things currently stand, around 17 cases of breast cancer are picked up in Scotland for every 2000 women who take part in routine screening - a service offered to all women aged 50 to 70.

But correctly identifying breast cancer is challenging, and roughly 20% of cases are missed in the screening process resulting in these patients being falsely given the all clear.

READ MORE: 

• Robots, AI, and health - will it cure us or kill us? 
• NHS Glasgow in 'world first' AI trial for COPD patients 
• Cancer screening and an ageing population: The dilemma for the NHS 

The Gemini initiative brought together Aberdeen University, Grampian health board, and Kheiron Medical Technologies, the software company behind 'Mia' - the AI tool trained to scour scans for breast cancer.

The aim was to evaluate whether adding Mia's expertise on top of the existing protocol (whereby every mammogram is read by two radiologists, and forwarded onto a third party if they disagree) led to better results.

In this sense then, the model actually added an extra step. However, the long-term goal is to bring Mia on board as the "second opinion" - meaning that in future mammograms could be examined by just one doctor plus Mia, with another human radiologist included in the process only if there was a lack of consensus.

Around 20% of breast cancer cases are missed during traditional screening (Image: Getty)

Dr Gerald Lip, who led the trial in Grampian, said this could reduce the average turnaround time for scan results from 14 to three days.

This is particularly important given the mounting demand for imaging, which is rising faster than almost any other area of healthcare.

The number of people in Scotland on waiting lists for CT and MRI scans, for example, has soared by nearly 40% over the past four years with close to 50,000 on the lists by the end of December.

Some of this is down to Covid-related backlogs, and some of it is due to the natural increase in demand from an ageing population.

Beyond this, however, shifting medical guidelines in the UK have also led to growing vigilance against conditions that are difficult to diagnose without imaging - such as blood clots on the lung - while overcrowded emergency departments are increasingly turning to scans as a triage tool to help medics weed out the patients who can be safely discharged from those who should be prioritised for admission.

Meanwhile, as patients with cancer survive longer they also require more regular and prolonged monitoring of how their body is responding to treatment - something which is usually done by CT scans, but occasionally also with more detailed PET scans.

An ageing population and more cancer patients surviving longer during treatment is increasing demand for imaging (Image: Getty)

In any case, the demand for imaging is only heading in one direction: up. Meanwhile, the workforce is struggling to cope.

In some parts of Scotland - Highland, Forth Valley and Fife - roughly 20-30% of consultant clinical radiologist jobs are vacant, and that doesn't even include the posts being filled on a temporary basis by locums.

According to the Royal College of Radiologists, Scotland is headed for a national 30% shortfall in radiologists by 2026 when the projected availability of medics is compared to the likely demand.

Which brings us back to the potential role of AI. Apart from freeing up staff, how much better did Mia perform compared to two human doctors?

According to the latest results of the evaluation, issued earlier this week, a total of 10,889 patients took part in the Gemini project while attending for their routine breast screening.

In line with national averages, just over 90 cases of breast cancer were successfully identified by the radiologists.

Breast cancer patients have a 95% survival rate if tumours are detected at less than 15mm (Image: Getty)

However, an additional 11 cases which had been missed by the doctors were detected by Mia - a 12% uplift.

This suggests that roughly 8% of patients are still slipping through the net, but - if rolled out Scotland-wide - it would nonetheless equate to some 220 extra breast cancers a year being picked up much earlier.

There was also evidence that Mia helped to reduce the false positive rate - the number of women being recalled unnecessarily for further investigations.

Barbara, an Aberdeen patient and one of the 11 women whose breast cancer was flagged by Mia but missed by the radiologists, was found to have a tiny 6mm tumour. It was so small that, following surgery to remove it, she only needed five days of radiotherapy.

READ MORE:

• Cancer patients in Scotland slower to start treatment than comparable countries 
• The all-in-one cancer blood test: A diagnostic Holy Grail, or a medical dead end?
• Scotland's 'woeful' cancer survival rates among the worst in West

While it is impossible to know how fast the disease would have progressed, most women now have a 95% survival rate if their tumour is detected when it is smaller than 15mm (half an inch).

Professor Lesley Anderson, chair of health data science at Aberdeen University, said the Gemini results "left us amazed".

She said: "If Mia were used in breast screening, it would mean that more cancers would be detected without putting more women through additional tests."

Dr Katharine Halliday, president of the Royal College of Radiologists, said the findings pointed to the use of AI as a "formidable force in patient care" for the future - albeit as an adjunct to human radiologists, not as a replacement for them.

The next step will see Mia rolled out in a much larger, nationwide NHS trial. If the results stack up, it could pave the way to the integration of AI into breast screening across the health service.

Sarah Kerruish, chief strategy officer at Kheiron Medical, added: "Just projecting these numbers across the UK, we're talking about thousands of women having their cancers - clinically meaningful cancers - picked up earlier.

"And when they're picked up earlier, they're easier to treat and the prognosis is much better.

"We're thrilled about that, but the real joy for us is being able to meet patients who had their cancers picked up by Mia.

"That's really what makes all of this worthwhile."