Adaptive personalization for wearable non-invasive continuous glucose sensors

Accurate continuous glucose monitoring (CGM) in non-invasive way have remained highly challenging, where current universal approach of “one-sensor-fits all” have been always frustrated, since the correlations between non-invasive parameters and blood glucose (BG) levels are highly individualized due to complex physiological states. For the first time to our knowledge, we proposed the methodology of personalized sensor-based “Adoptive Non-invasive CGM” that could potentially achieve the “holy grail” of accurate CGM in non-invasive way. The device included a short-term used microneedle minimally-invasive CGM (MI-CGM) module to measure BG in interstitial fluid, and a long-term used non-invasive CGM (NI-CGM) module based on metabolic heat conformation. The NI-CGM module of the worn device on body gradually (~2 days) learned the individual’s physiological characteristics and the accurate glucose sensing capability from MI-CGM module through GRU-based AI model, ultimately forming personalized sensor tailored for each unique patient. The MI-CGM module was then removed while only the NI-CGM module was remained for longer-term use. The short-term (5-consecutive days) performance of Adoptive Non-invasive CGM was demonstrated on 6 participates, with clinical-acceptable accuracy (MARD~15.4%). This method also presented good reproducibility in parallel testing, and possessed reasonable accuracy (error ~20%) during long-term (>2 months) use. This was likely the first time that non-invasive CGM was demonstrated to meet clinical standards of accuracy (MARD≈15%) in multi-day continuous monitoring, while BG trends also highly match the actual BG curve fluctuation. Our personalized sensors could potentially change the conventionally used strategy of developing CGM to adapt universal patients, and held promise for addressing the “holy grail” problem in CGM field.