AI is transforming how radiologists detect breast cancers
A resource-strapped NHS and the deep-learning community should make the perfect bedfellows. One is a public-sector behemoth sitting on a treasure trove of patient data. The other features entrepreneurs backed by algorithms and missionary zeal.
However, concerns around patient confidentiality and clinical success rates have led some organisations to adopt a more cautious approach.
The healthcare start-up Kheiron Medical Technologies wants machine learning not just to help doctors find breast cancer earlier, but also more accurately.
One woman in eight will develop breast cancer in their lifetime. Currently, women aged 50 to 70 are invited to have an NHS screening every three years. Each mammogram is double-read by two independent clinicians before a decision is made about whether to call the woman back for further investigation.
Because of a shortage of radiologists, particularly in breast imaging, backlogs have formed. Tumours can be missed and overdiagnosis is not uncommon. The trauma of recalling women for biopsies – sometimes unnecessarily – complicates the matter.
Such a situation appears ripe for artificial intelligence (AI). Founded in 2016, Kheiron has been quietly going about its business – until now. In July it became the first UK recipient of a European Union CE health safety mark for deep-learning applied to radiology.
The CE mark rubber-stamps the start-up’s rigorous methodology. The company is already working with global and UK partners, including the East Midlands Radiology consortium, which covers eight NHS trusts, to assess its algorithm’s performance test trials.
Kheiron has developed deep-learning software to act as a second reader of mammograms, replacing one of the clinicians and potentially halving the workload of overstretched staff.
“CE marking for us is not a ticket to start selling [the product] and printing money. There has been no push for marketing, hype or sales,” says Dr Hugh Harvey, a clinical director at Kheiron and a practising radiologist who sits on the Royal College of Radiologists AI working group.
Dr Harvey formally joined Kheiron in January after its founders – Dr Peter Kecskemethy and Tobias Rijken – outlined their calm, step-by-step approach. He believes we’re now “four or five years” from a recommendation by the National Institute for Health and Care Excellence (Nice) for introducing deep-learning algorithms for radiology.
Sarah Kerruish joined Kheiron after working in California’s Silicon Valley for 15 years, where she was something of a force of nature. Aside from her experience in the healthcare sectors of both the US and the UK, Kheiron’s chief strategy officer is also an Emmy-nominated director – for the documentary Moon Shot, about the Nasa space programme.
Kerruish says that the UK “has an extraordinary opportunity to be a global leader in leveraging deep learning to support doctors and deliver better outcomes to patients, including detecting more cancers at the crucial early stages”.
Indeed, after Silicon Valley, London is becoming the hotbed for AI. DeepMind, Google’s AI company, is probably the best known. In partnership with Moorfields Eye Hospital in London it has developed a system that can detect 50 features of eye disease in seconds, with 94 per cent accuracy – matching world-leading eye experts.
Dr Harvey says that news outlets often assume that AI is a silver bullet, which will replace doctors or surgeons. Rather, it will allow them to do ever more complex tasks and offer care that machines can’t do.
“That is the real story,” he says. “But that’s not as sexy.”