WSEAS Transactions on Information Science and Applications
Print ISSN: 1790-0832, E-ISSN: 2224-3402
Volume 9, 2012
Neural Network Based Filter for Continuous Glucose Monitoring: OnTuning with Extended Kalman Filter Algorithm
Authors: ,
Abstract: This paper deals with removal of errors due to various noise distributions in continuous glucose monitoring (CGM) sensor data. A feed forward neural network is trained with Extended Kalman Filter (EKF) algorithm to nullify the effects of white Gaussian, exponential and Laplace noise distributions in CGM time series. The process and measurement noise covariance values incoming signal. This approach answers for the inter person and intra person variability of blood glucose profiles. The neural network updates its parameters in accordance with signal to noise ratio of the incoming signal. The methodology is being tested in simulated data with Monte Carlo and 20 real patient data set. The performance of the proposed system is analyzed with root mean square(RMSE) as metric and has been compared with previous approaches in terms of time lag and smoothness relative gain(SRG). The new mechanism shows promising results which enables the application of CGM signal further to systems like Hypo Glycemic alert generation and input to artificial pancreas.
Search Articles
Keywords: Continuous Glucose Monitoring, Denoising, Extended Kalman Filtering, Laplace noise, Neural network, RMSE