Adaptive personalization for wearable non-invasive continuous glucose sensors

Accurate continuous glucose monitoring (CGM) is essential for diabetes management, yet non-invasive CGM remains challenging. Universal "one-sensor-fits-all" approaches often fail, because correlations between non-invasive physiological readouts and blood glucose (BG) are highly individualized. Here, we propose a personalized "Adaptive Non-invasive CGM" methodology, which integrates a short-term minimally invasive CGM (MI-CGM) module for measuring interstitial fluid glucose, and a long-term non-invasive CGM (NI-CGM) module based on metabolic heat conformation. During a ~2-days training, the NI-CGM learns accurate sensing from the MI-CGM via GRU-based AI model, forming patient-specific sensor. The MI-CGM is then removed, leaving only the NI-CGM for long-term monitoring. In a 5-day CGM test, the system achieved MARD ~15.6%. This method also demonstrated good reproducibility and maintained reasonable accuracy (MARD ~23.9%) during long-term interval use (>2 months). This personalized approach offers a pathway from conventional "one-sensor-fits-all" strategies toward individualized sensor development, potentially mitigating inter-patient variability and non-invasive monitoring inaccuracies.