3 myths about translating AI models in a healthcare setting

3 myths about translating AI models in a healthcare setting

Artificial intelligence has huge potential for many healthcare challenges – but there are still many hurdles that must be overcome.  

At a keynote for the Association for Computing Machinery Conference on Health, Inference and Learning this past week, Dr. Alan Karthikesalingam, research lead at Google Health UK, described three myths commonly encountered in the path to building and translating AI models in clinical settings.  

When it comes to implementing deep learning technology, he asked: “Why is there a gap between expectations and reality?”   

Here are three common misconceptions Karthikesalingam said must be addressed.  

1. More data is all you need for a better model.  

The problem, he said, is that what we might regard as “ground truth” is more subjective than we think. One ophthalmologist might look at images of an eye and see moderate degeneration, whereas another would see it as mild.  

“Doctors don’t always agree,” Karthikesalingam explained.  

The quality of labels seems to make a big difference in this regard.

“Choosing an efficient labeling strategy” is one way to ensure quality, he said, “but also taking other modeling approaches and bringing them to bear.”  

2. An accurate model is all you need for a useful product.  

On the contrary: a human-centered approach is key to building useful products.   

Karthikesalingam’s team found that creating AI “onboarding” changed their understanding about what users need from tools. 

“Product usability is incredibly important, and comprises a whole raft of other kinds of activity around which model development has to adjust,” he said.  

3. A good product alone is sufficient for clinical impact.  

“Post-market, careful independent study takes a long time,” said Karthikesalingam.  

“Implementation and health economic research are critical to adoption of AI products,” he added.

Overall, examples of deep learning are all around us as consumers – and the medical field will eventually be no exception, said Karthikesalingam  

“Technology, when it works well, should make it as easy as possible” to treat patients, he said.

Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: [email protected]
Healthcare IT News is a HIMSS Media publication.

Source: Read Full Article