BpMC: A novel algorithm retrieving multilayered tissue bio-optical properties for non-invasive blood glucose measurement

摘要

Non-invasive blood glucose measurement is a crucial challenge in both academic and industry communities. Currently, most of non-invasive solutions are developed based on optical signals. However, their accuracy is still far from clinical requirements if these measured optical signals directly used to estimate corresponding glucose levels. To solve this challenge, a novel Back-propagation Monte Carlo (BpMC) algorithm is proposed to retrieve bio-optical properties in human multilayered tissues. Build on BpMC algorithm, two non-invasive blood glucose estimation models, namely BpMC-DEE and BpMC-CNN, are conceived. In contrast to existing black-box solutions, BpMC-DEE is a white-box model that is more reliable in clinical. BpMC-CNN is a gray-box model whose results are more accurate in cost of a larger dataset and higher computing complexity. BpMC-DEE and BpMC-CNN are embedded and implemented into our designed noninvasive device - Earlight, for clinical trials. The clinical trial results demonstrate that correlation coefficients of these two models reach 0.852 and 0.895, respectively, referring to invasive glucometers. In terms of Clarke Error Grids, our proposals account for 90.6% and 93.5% statistic points in regions A and B, respectively. Moreover, the BpMC algorithm can be applied to other components measurement of biological tissues.