AI in Diabetes Management

Diabetes management revolves around keeping glucose within a target range. Continuous glucose monitors (CGMs), such as those from Dexcom and Abbott's FreeStyle Libre line, measure glucose in the fluid under the skin every few minutes, producing a data stream well suited to prediction. Many CGM systems show trend arrows and predictive alerts that warn of an impending low or high. The biggest application is automated insulin delivery (AID), often called an artificial pancreas. An algorithm reads CGM data and adjusts an insulin pump's delivery. The FDA approved Medtronic's MiniMed 670G, the first hybrid closed-loop system, in 2016. Later systems include Tandem's Control-IQ, Insulet's Omnipod 5 and Beta Bionics' iLet, which starts dosing from body weight rather than traditional pump settings. 'Hybrid' means users generally still announce meals or give meal boluses, because insulin acts more slowly than food raises glucose. A community open-source movement, including OpenAPS and Loop, built do-it-yourself systems before and alongside commercial ones. Most AID algorithms are not deep learning. They are control algorithms, such as model predictive control or proportional-integral-derivative control, often with adaptive features that adjust to a user's needs over time. Machine learning is more prominent in glucose forecasting research and decision-support apps. In eye care, IDx-DR (now called LumineticsCore, from Digital Diagnostics) received FDA De Novo authorization in 2018 as the first autonomous AI diagnostic system, detecting more-than-mild diabetic retinopathy without a clinician interpreting the image. Other systems, including Eyenuk's EyeArt, followed, and Google has studied retinopathy screening in India and Thailand. A common misconception is that closed-loop systems cure diabetes or make it fully automatic. They reduce burden and improve time in range for many users, but still require sensor changes, meal input and attention to device problems.