3 myths about translating AI fashions in a healthcare setting

Synthetic intelligence has big potential for a lot of healthcare challenges – however there are nonetheless many hurdles that have to be overcome.  

At a keynote for the Affiliation for Computing Equipment Convention on Well being, Inference and Studying this previous week, Dr. Alan Karthikesalingam, analysis lead at Google Well being UK, described three myths generally encountered within the path to constructing and translating AI fashions in scientific settings.  

With regards to implementing deep studying know-how, he requested: “Why is there a niche between expectations and actuality?”   

Listed here are three frequent misconceptions Karthikesalingam mentioned have to be addressed.  

1. Extra knowledge is all you want for a greater mannequin.  

The issue, he mentioned, is that what we’d regard as “floor reality” is extra subjective than we expect. One ophthalmologist would possibly have a look at photos of a watch and see reasonable degeneration, whereas one other would see it as delicate.  

“Medical doctors do not at all times agree,” Karthikesalingam defined.  

The standard of labels appears to make a giant distinction on this regard.

“Selecting an environment friendly labeling technique” is a method to make sure high quality, he mentioned, “but in addition taking different modeling approaches and bringing