When asked what is the most important lesson he’s learnt in his artificial intelligence journey, Haris Shuaib, Senior Physicist in Magnetic Resonance and Topol Digital Fellow at Guy’s and St Thomas’ National Health Service (NHS) Foundation Trust stated it was the value of independent testing. “Most people know AI performance fluctuates from one setting to the other and end-results can be radically different,” he says. “So, it is important to validate AI-driven products and do that as early as possible. If you are working with technology partners, get them to commit or facilitate that testing”.

Shuaib believes AI validation does have some impact on the widespread AI deployment as the biggest barrier right now is to convince the majority that the technology can do something different or have a significant impact in medicine, particularly radiology. “We already have many market-approved AI products so the primary issue is not the availability of products. What we need is to test the products that are available on the market. There is not enough evidence to prove that AI can lead to systemic benefit or large scale benefit to patients at the moment. I think they are due to be established”.

People fall out of love with a technology because it does not work

Nevertheless, even if it becomes apparent that AI does benefit our healthcare system. Shuaib warned the danger of becoming overconfident since both medicine and technology evolve in a rapid manner. Technology partners should continue to spare thoughts on workflow integration and user experience as they upgrade AI products.

“People fall out of love with electronic health records (EHRs) because it is an outdated software that does not serve the purposes it was designed for. It is making clinicians’ lives even more difficult,” Shuaib says. “But AI is different from EHRs in the sense that AI is targeted at very specific clinical problems, whereas EHRs is a system that is not developed for any specialty. Likewise, we cannot compare telemedicine with AI because no one ever doubt the usefulness of telemedicine. It was investment which resulted adoption delay”.

“While AI also has its investment problems, there is no concrete evidence, as I mentioned early on, to indicate that AI helps the way we do medicine. It only helps us with very specific tasks. Either or, as far as I know, clinicians will remain excited about things that help them to help their patients. I have not experienced and I would not expect clinicians losing their interests in a successful technology that is doing well. So, if that excitement ever wanes, it is a very good indicator that something must be wrong with the product”.

Same level of commitment is required throughout the entire AI life cycle

On top of making sure that AI works as intended. Shuaib adds commitment is equally important in keeping clinicians’ and stakeholders’ interests in AI. “This is a difficult and long protracted process. We have to effectively convey the benefits of deploying AI in the institution or department and sustain that commitment thereafter. It is important to ask, not just on Day Zero, but along the way, ‘Do you still have that same commitment?’ because challenges are not always at the beginning, they are throughout the AI life cycle. So, maintaining a good level of communication and appropriate engagement with all the different stakeholders is crucial”.

As such, Shuaib advice clinicians who starting their AI journey to focus on the problem, rather than developing AI itself. Besides, think about the business side of things such as procurement mechanisms; agile deployment, post-market monitoring, decommissions and so on. Shuaib thought hospitals or healthcare institutions should be taking the lead in AI deployment and there ought to be incentives.

“I think the barriers to widespread AI deployment has to be tackled by healthcare providers. People who work in medicine and healthcare, their primary drive is to improve operations for the benefits of patients and the institutions. So, healthcare leaders need to essentially smooth the pipeline from creation to bedside and give feedback to technology partners so that they can develop better products.

So, the power is really in our hands; lead within the system, make a more concerted and directed effort in various investments; from training staff to validate AI, curate large and appropriate datasets, to providing avenues for developers to train AI models in agile environment. There is no need for mega projects but quick trials and that would be the most beneficial”.

Haris Shuaib will be joined by fellow clinicians, healthcare leaders, C-suite executives, technical experts and many more in the upcoming AIMed Radiology virtual conference in association with the American College of Radiology on 5 November. You may register your interest or obtain a copy of the event agenda here.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.