Almost 30,000 women a year in the UK could have their lives saved or extended by a new AI breast cancer diagnosis tool that could be available on the NHS within three years, developers say.
Researchers from King’s College London are working with doctors at Guy’s and St Thomas’ hospitals and Google-backed startup Owkin to finalise and trial the new device.
it is being trained to spot a sub-category of breast cancer that is not picked up with conventional diagnostic tests but could be treated much more effectively if it was.
This kind of cancer, known as HER2-low, is thought to be present in about half of the 55,000 new cases of breast cancer that emerge in the UK each year.
The early signs are that an effective new drug for another form of breast cancer could also work well against HER2-low – although more research is needed to confirm this.
“This has the potential to radically improve outcomes for patients across the NHS and beyond,” said Thomas Clozel, co-founder and chief executive of Owkin.
“Research suggests that many more women could benefit from targeted breast cancer treatments – we just need to find them. We hope to help thousands more women to benefit from targeted anti-HER2 treatments in the UK every year, with the transformative drugs able to extend and save patients’ lives.”
The HER2 proteins behind this form of cancer are overproduced by faulty genes. The new AI models will analyse hundreds of retrospective tissue samples, meaning that the amount of HER2 proteins present in the cancer can be determined more quickly and accurately from biopsy samples.
The developers of the device stress it is still very early days but that the indications are that their AI model will prove to be effective at picking up cases of HER2-low cancer, and that patients will, in turn, benefit from new drugs.
Dr Sheeba Irshad, breast cancer medical oncologist at Guy’s and St Thomas’, said: “If these methods prove to be successful, use of AI-aided assessment of tumour tissue could be rolled out to more quickly and accurately screen patients for these markers.”
Experts not directly involved in the project were also excited by its potential – again cautioning that much more work is needed to confirm its potential.
“The new AI models hold great promise for extending the benefits of targeted treatments to more women, especially those with lower levels of HER2 proteins,” said Dr Nigel Blackburn, director of drug development at Cancer Research UK, speaking in a personal capacity.
Dr Jakob Nikolas Kather, assistant professor at the University of Aachen and visiting professor at the University of Leeds, said: “This could speed up the detection of aggressive subtypes of cancer and help more women to receive optimal treatment.”
What is HER2 cancer and how can it be treated?
The HER2 proteins behind HER2-low cancer are overproduced by faulty genes. In larger quantities, they produce a particularly aggressive form of breast cancer known as HER2-positive, which affects about 8,000 women in the UK a year.
This can typically – although by no means always – be picked up by standard diagnostic tests.
HER2-low, meanwhile, is a sub-category of the cancer, involving smaller amounts of the protein that do not show up in existing tests.
Apart from spotting these cases for the first time, the researchers are confident their device will be quicker and more accurate at diagnosing standard cases of HER2. While typically spotted, these are misdiagnosed in about 20 per cent of cases, previous research has shown.
Dr Kotryna Temcinaite, of charity Breast Cancer Now, who is not involved in the project, said: “If we look more closely at breast cancers that are currently classed as HER2 negative, some could be called HER2-low. While HER2-low is not currently defined as a type of breast cancer in its own right, it is starting to be considered as a subtype and estimations suggest that around half of breast cancer cases may qualify as HER2-low.
“The benefit of targeting the HER2 proteins in treatment of breast cancer that is HER2-low has not been confirmed in clinical trials yet. However, some promising early data from laboratory research suggests some newer targeted drugs developed to treat HER2 positive breast cancer could also be used to treat people with HER2-low breast cancers.”
A number of drugs used to treat HER2-positive cancer could potentially work well against HER2-low, with AstraZeneca’s new trastuzumab deruxtecan treatment, sold under the brand name Enhertu, causing particular excitement. On Monday it was approved for use in Scotland, following recent approval in England, Wales and Northern Ireland.
Trastuzumab deruxtecan combines two drugs and attaches itself to the HER2 proteins which can stop cancer cells growing. When the trastuzumab attaches itself to the proteins, it delivers anti-body drug deruxtecan directly into the breast cancer cells to kill them.
A study published in September found the drug reduced the risk of death or disease progression by 72 per cent compared to an existing treatment, AstraZeneca has announced.
Three-quarters of patients showed no progression in their disease after 12 months compared to 34.1 per cent of those treated with trastuzumab emtansine (T-DM1), while progression-free survival was improved from 7.2 months to 25.1 months.
Nearly all Enhertu patients were alive after 12 months (94.1 per cent), compared to 85.9 per cent of T-DM1 patients.
How artificial intelligence is transforming healthcare:
The new breast cancer diagnostic device is part of a new generation of artificial intelligence tools with the power to transform cancer treatment and diagnosis and the healthcare system more broadly.
The developers of the HER2 tool are planning to use it to better diagnose gastric cancer further down the line, as this form of the disease can also involve an overexpression of HER2 proteins.
HER2 proteins are not found in other kinds of cancer so there would not be any further applications for this particular tool after that.
But AI is already being used against cancer in a range of applications.
Owkin, the French company behind the HER2 device, has previously built effective AI models for mesothelioma and colon cancers and to predict strokes and heart attacks.
In October, researchers from the Institute of Cancer Research (ICR) and The Royal Marsden NHS Foundation Trust launched a major study to analyse up to 5,000 tumour samples in minute detail using artificial intelligence.
They hope thousands of lives could be saved through the new tool, which is expected to revolutionise diagnosis and treatment of a dangerous form of cancer known as sarcoma.
And more than 22,000 cases of lung cancer missed every year in X-ray scans could be picked up by a new AI lung cancer detection system being trialled by the NHS on 150,000 patients at hospitals in the Southampton and Dartford areas.
