The Oregon Health & Science University (OHSU) says it has achieved success in using an AI system to treat people with Type-1 diabetes. According to the team behind this effort (you can read their report at Nature Metabolism), the AI’s suggestions, given to patients via an app, were in agreement with doctors’ recommendations 67.9% of the time, with over 99% of these meeting safety standards. This could potentially help patients who are unable to seek medical advice at recommended intervals, and with untreated diabetes often leading to severe complications, could even prove to be lifesaving. While 67.9% might not sound that much, the authors point out that the rate of agreement between physicians’ recommendations is around 41% to 45%. As lead author Nichole Tyler explained, “We designed the AI algorithm entirely using a mathematical simulator, and yet when the algorithm was validated on real-world data from people with Type 1 diabetes at OHSU, it generated recommendations that were highly similar to recommendations from endocrinologists.”
AI-powered smartphone apps could help early diagnosis
Diabetes isn’t the only illness where AI has been proven to make a difference, and as more data gets collected and algorithms become smarter, expect many more such apps to come up. The OHSU’s Mole Mapper app might not have a sizable enough dataset to offer diagnosis, but allows people to track the growth of moles over time. Meanwhile, a team at Stanford says its AI system can match – and even outperform – dermatologists in identifying and classifying skin cancers from photographs. But perhaps the most famous app of this type is SkinVision, which claims to detect 95% of skin cancers. Other apps exist to diagnose illnesses that affect the eyes, for example, Baylor University’s Cradle app, which was able to detect leukocoria, or ‘white eye’ (which can often signify the presence of serious disorders) in 80% of affected children.
(Photo Credit: Alexander Sinn)
