MedTech data should represent entire population when training AI, says Department of Health director

When training AI models, MedTech companies must ensure that the data being used is representative of a population group in its entirety, according to David Lawson, director of medical technology & innovation at the Department of Health and Social Care.

Speaking to National Technology News at an event by Startup Coalition for MedTech companies at the House of Lords, Lawson said companies need to make sure that the data incorporates different ethnic groups.

“There are risks around using data when you are designing medical devices, particularly when training AI technology, to ensure you are not factoring in potential bias,” he said. “Men and women respond differently, as do ethnic minorities.”

Lawson added that both patients and clinicians need to be able to trust the devices and solutions on offer, and this means having confidence in the data.

AI maturity across medical disciplines

Elsewhere, Lawson explained that certain medical disciplines are more prepared to adopt new technologies as they are more technically advanced.

Lawson said that neurology, which is using digital imaging, is more likely to adopt AI than pathology, as the latter still relies on non-digital processes.

“You can’t apply AI to paper-based systems – you need that base digital infrastructure for it to be embedded,” he said.

The medical director added that there is also a limit to how much change a healthcare system such as the NHS can take on, claiming that this is sometimes forgotten by MedTech companies.

“If a health system is going to adopt something, we have got to make sure they also have the resources to implement it,” Lawson explained. “The technology is the easy part – it’s the time to be trained on a new piece of kit or process, plus the follow up to ensure that the change is embedded and compliant.”

Lawson went on to say that the majority of MedTech is purchased on a local level, with some firms finding themselves “knocking on every door” to persuade hospitals to adopt their technology.

A standardised process

He said that the health department is working on a standardised process for making decisions, meaning that if a technology is proven to work in one place it can be adopted in multiple locations without a lengthy evaluation process.

In terms of MedTech that could make an impact in the coming years, Lawson called the pace of development around AI technology “dramatic” and said it is going to increase.

Lawson added that quantum sensing technology is another area of growth, while engineering technology has the potential to create artificial blood.

“One technology is using targeted soundwaves to treat liver cancer in a non-invasive way,” Lawson added. “In future years as they do more clinical work on other cancer tumours, the ability to treat cancer in a non-invasive way is game changing both in terms of the ability to provide cancer treatment in a community setting without certain side effects.”