‘AI is an enhanced search engine’
GEHealthCare spun out from its massive parent company General Electric earlier this year. When the plan was announced in 2021, the company said GE HealthCare’s focus would be on driving innovation in precision health to address challenges facing patients and clinicians.
Even before the spinoff, GE’s healthcare division had been investing in artificial intelligence and imaging technology for years. Parminder Bhatia, GE HealthCare’s chief AI officer, joined the team in April after spending time at Amazon and Microsoft. He leads a group of employees that have brought 13 AI algorithms to market in the past year.
In an interview, Bhatia spoke about how GE HealthCare’s strategy has evolved beyond medical imaging equipment and his predictions for the future of AI in healthcare.
GE HealthCare has developed algorithms whose use cases include predicting missed care incidents and guiding ultrasound interpretations. What do you say to those who are worried about AI tools replacing the work of clinicians, or about AI in general?
There are questions … on data, cybersecurity and how these models are built.
Our customers and partners feel confident when the models are validated at multiple sites.
One of the biggest challenges we hear is, “This is a black box that gave me a recommendation.” But if we can provide the reasoning—this model gave [this output] because of these guidelines— then [the technology] actually becomes like a superpower to our clinicians or radiologists, allowing them to make decisions at a high level. All it’s trying to do is automate routine, mundane tasks. That way, [clinicians] actually spend more time with the patients.
The way I think about AI in healthcare, in general, is: “How can this be an intelligent companion?” AI is an enhanced search engine that provides radiologists and clinicians with the right tools so that they can make the right [decisions]. The [decision] is still made by the radiologist or clinician. The tool is just providing them with or resurfacing the right information, so that they can make judgments based on those components as well.
Post-pandemic, we’ve seen a lot of focus on the operational and clinical efficiency side of [AI tool development].
How reliable do these algorithms need to be before you feel ready to release them on the market?
When we’re building these models or capabilities, we believe
in rigorous multi-site validation. … We evaluate these models across multiple locations with our partners before we think they are ready for real-world application.
A lot of these components have a human in the loop, such as a radiologist. We’re going to provide them with the [guidance on] what the output is and why that is coming. … So different capabilities would have different levels of accuracy,
I would say. There is no one hard rule that it needs to be 90% or 95% effective.
Does GE HealthCare have goals to become a software company, rather than a company largely focused on medical imaging devices?
We think of it as a “D-3” strategy: smarter devices, disease-specific focus and digital solutions. When we talk about these digital solutions, some of them would be tied to the devices and some of them could be built in [to other companies’ devices] as a multi-vendor approach as well. Any of the disease care pathways would see a multi-vendor approach.
We’ve made huge investments in software or digital applications that can help our customers solve problems of operational efficiency, as well as how they can reduce overall costs and see impacts across different disease care pathways.
How much of the building that you’re doing relies on out-of-the-box large language models from software companies like OpenAI, Microsoft, Google or Amazon, versus your own?
I would say it’s a hybrid approach. For some capabilities—for instance, anything on medical imaging—that’s where we are building the foundation models. [When we use others’] large language models, we can also think about doing strong partnerships—to really double down some of these efforts to build these capabilities together.
Another approach involves an acquisition. For example, GE HealthCare acquired [ultrasound technology company] Caption Health
n earlier this year.