The Leeds doctor at the forefront of trialling artificial intelligence in NHS breast screening service

“It’s important because it enables human readers (of mammograms) to interact with AI for the first time,” she says. “The potential is really exciting. When we have the opportunity to do the trial and interact with AI, I think we’ll start thinking more how can we handle things differently.”

The need for change is pressing. The latest Royal College of Radiologists’ clinical workforce census found the UK now has a 29 per cent shortfall of clinical radiologists, which will rise to 40 per cent in five years without action. By 2027, it said an additional 3,365 clinical radiologists would be needed to keep up with demand for services and concerns have been raised about longer waiting times and worse health outcomes.

In 2021-22, the NHS invited record numbers of women to attend breast screening appointments – the highest ever for a single year. “We aren’t going to have enough radiologists to run the screening service,” says Dr Sharma, “and therefore we need to start thinking innovatively about how we can work differently and maintain quality and safety. And I do think artificial intelligence can help us in that way.”

Dr Sharma is at the forefront of trialling artificial intelligence and is working with tech companies to make sure their AI addresses clinical needs. She wants to find creative solutions to the NHS workforce challenges, whilst ensuring patients receive the best possible care.

The LIBRA study, funded as part of NHS England’s £100m AI in Health and Care Award, will see the Leeds Teaching Hospitals NHS Trust deploy an AI reader to work alongside two human readers to analyse mammograms for tell-tale signs of cancer.

Under the current standard of care, every mammogram is seen by two human readers. If they disagree on their readings, a third reader will review the mammogram film in a process known as arbitration before a decision is made whether to recall the patient or not.

In the LIBRA study, if the two humans and the AI reader agree a mammogram is normal, the patient will receive the all-clear. If any of the human or AI readers disagree, clinicians will carry out the arbitration before making a decision on recall.

The study is designed to be safe and non-invasive and if successful, it could lead to an AI reader replacing one of the two human readers in the standard model of care. This could free up clinicians to spend more time with patients, speed up screening and increase the numbers of women seen by the service, the trust said.

“When you have new technology coming to the forefront, everyone is going to be anxious, not only the patients but clinicians,” Glasgow-born Dr Sharma, the chief investigator, says. "We’re testing the algorithm in a safe environment to see if it can deliver our clinical expectations. If it can, we’ll deploy it in a clinical setting but the public can be reassured that we have evidence to say it’s safe to use. And clinicians will have that reassurance too.”

Dr Sharma has also been working on another project looking at how AI can assess the quality of the mammograms and provide real-time feedback to mammographers. The quality is important to help identify cancers on the mammogram. “If the image you’re reporting on isn’t of good quality, then it’s going to be challenging for the reader to identify the cancer, whether it’s a human reader or artificial intelligence,” Dr Sharma says.

Ultimately, it is hoped that AI will lead to better quality images, helping to identify additional cancers and improving health outcomes of patients in Leeds as a result. Dr Sharma says: “AI works differently to humans and by combining human expertise with technology advances this could lead to earlier detection of cancers and impact positively on women’s lives.”