A Dietary Feature-Integrated Non-Invasive Corneal Autofluorescence System for Systemic AGE Assessment

Current clinical indicators for diabetes management are limited by the invasiveness of conventional measurements and their inability to capture metabolic memory, highlighting the need for reliable biomarkers for long-term metabolic status assessment. Advanced glycation end products (AGEs) accumulate progressively in corneal tissues. In this study, a portable AGE detection system based on corneal autofluorescence was developed. To enhance the acquisition capability of weak fluorescence signals, the system integrated a silicon photomultiplier (SiPM) with a dual-phase lock-in amplification technique. Multiple regression models were established using clinical data from 300 participants, with dietary factors incorporated as additional input features. Among them, serum AGE measurements from 104 participants were used as reference values. The Random Forest Regression model exhibited the best predictive performance, with approximately 85% of samples showing absolute relative prediction errors below 5%, a mean absolute percentage error of 3.8%, and a maximum relative error of 12.01%. Longitudinal monitoring further demonstrated that prediction errors were below 7% at three of the four time points, while the error at Day 0 was 10.77%. These findings suggest that the proposed system can provide relatively accurate and stable estimates of serum AGE concentrations and may have potential for supporting for long-term diabetes monitoring and early screening.