Non-invasive, Intelligent Blood Glucose Monitoring on Fingertip Using Dual-band Fusion and LSTM-R Network

Non-invasive, painless, and continuous monitoring sensing systems are needed to enhance the life quality of the diabetic. In this paper, two finger-sized ultra-wideband (UWB) antennas are used as detectors of blood glucose level (BGL) sensing system. A dual-band fusion strategy and an improved long-short-term memory (LSTM-R) are introduced based on UWB microwave technique, to overcome the negative effects of antenna miniaturization. The proposed composite end-to-end framework can achieve the high accuracy and robustness of BGL prediction. The safety and reliability of proposed microwave-based sensor were verified by simulation and experiment. The reported sensing system is validated on the medium under tests (MUT) with different glucose solutions and the human trials. For the measurement of MUT, the testing results achieve a root mean squared error (RMSE) of 4.2367 mg/dL and a mean absolute relative difference (MARD) of 2.79%. During human trials, the non-invasive estimation results obtain a low MARD of 5.9%, and are fallen 100% into the clinically acceptable zones by Clark error grid analysis.