Blood glucose estimation using third person continuous glucose monitoring data via complete ensemble empirical mode decomposition with adaptive noise and long-short-term memory networks

Abstract

Background Blood glucose estimation is critical for monitoring the health of patients with diabetes. Traditionally, it has been performed invasively, via pricking a fingertip to draw a small amount of blood for analysis. However, this type of approach causes pain to patients in the long-term. More recently, micro-invasive approaches have been proposed as alternatives; however, these new methods can be quite costly. Methods To address this issue, we proposed a minimally invasive approach for obtaining blood glucose measurements. In our proposed approach, the continuous glucose monitoring data of a third person are mapped to more sparse data from a particular user, acquired through traditional invasive means. The complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method is then applied to the mapped glucose data. The first six intrinsic mode functions are discarded, and a long short-term memory (LSTM) network is used to perform non-invasive blood glucose estimation. To demonstrate the effectiveness of our proposed method, a loss function was used as a performance metric. Our approach was compared to the LSTM network, both with and without an empirical mode decomposition (EMD) approach. Results Numerical simulation results showed that our approach achieved a higher accuracy than either of these two popular methods. Conclusions Our method can be used to improve the accuracy of blood glucose prediction.