Somerset trust cuts cancer referral pathway with AI

Published: 15-Jun-2022

Somerset NHS Foundation Trust halves the time through a key stage in the lung cancer referral pathway in a validation study of diagnostic AI

Somerset NHS Foundation Trust has become the first in the UK to study the performance of a pioneering AI algorithm for detecting lung cancer from X-rays.

The trust has been testing whether the award-winning Behold.AI can help it meet national targets for lung cancer diagnosis.

And its initial results suggest red dot has played a part in helping to more than halve the time from initial X-ray screening to a CT scan, the gold standard for detecting lung cancer.

“There’s been a lot of buzz about AI at radiology meetings, but there’s little experience of using it in an NHS trust,” explains Dr Paul Burn, consultant radiologist at the trust.

“We embarked on a bottom-up initiative to test the algorithm, with the aim of helping us improve our referral times.”

He added: “We have a fairly-elderly patient population, which may make it harder for AI imaging solutions to be effective because of a higher incidence of abnormalities that show up on X-rays, such as scarring and calcifications.”

But, by prioritising which X-rays need urgent attention from a radiologist, the AI helped reduce the time from chest X-ray to CT scan from seven to 2.8 days, supporting the trust to meet its 28-day cancer diagnosis target.

As part of the study, the trust also adjusted its next-day booking pathway for CT scans, the next stage in the lung cancer referral process after a chest X-ray.

The trust and Behold.AI found that of the 3,794 chest X-rays reviewed by the red dot algorithm over a three-month period, the average time for a result to enter the hospital systems was 16 seconds.

The algorithm, which was developed in collaboration with NHS consultant radiologists, provides two outputs – a subset of abnormal X-rays with a high probability of lung cancer, and another subset of X-rays (high confidence normal) with a very high likelihood of being normal.

Of the 3,794 chest exams, the red dot service classified 562 (14.8%) as high confidence normal (HCN).

In 13 cases, radiologists disagreed with the model’s classification as HCN, a negative predictive value (NPV) of 97.7%. None of these discrepancies were considered clinically significant.

“High Confidence Normal results are an obvious opportunity for where AI can be used in the future,” said Dr Burn.

“Particularly for trusts with a big backlog reporting problem.’’

A study carried out by Somerset NHS Foundation Trust showed the time taken from chest X-ray to CT scan was reduced from seven to 2.8 days as a result of using the AI algorithm

A study carried out by Somerset NHS Foundation Trust showed the time taken from chest X-ray to CT scan was reduced from seven to 2.8 days as a result of using the AI algorithm

Pioneering AI start-up, Behold.AI, won Best UK Digital Health Solution for red dot at the Prix Galien Awards ceremony on 12 May.

Simon Rasalingham, company chief executive and chairman, said: “When we talk about the next big problem after COVID, it’s going to be the NHS backlog, which will reach 12 million patients, nearly a quarter of the adult population of the UK.

“The peer-reviewed evidence we’re building up shows that if we roll this technology out nationally we can add in the equivalent of 233 NHS consultant radiologists, which is equivalent to 255,000 hours of consultant time per year.”

He added: “Early-stage lung cancers are often missed by X-rays and we believe that our technology can pick up 22,000 more cases of lung cancer every year, giving these people a significantly-better chance of beating the disease.”

Dr Michael Marsh, South West medical director for NHS England and NHS Improvement, said: “The pioneering work carried out by the radiology team at Somerset NHS Foundation Trust to test the AI technology in a hospital setting is a great example of innovation taking place in the South West NHS to reduced waiting times and improve outcomes for patients.”

You may also like