Denoising of photoplethysmograms for non-invasive blood glucose estimation via Slant transform based bit plane method

Abstract

It is worth noting that the diabetes is a chronic disease. It causes an irreversible damage to the human organs and even leads to the death. Moreover, for the mortality due to the COVID19, it was found that the death rate of the subjects with the diabetes is higher than that without the diabetes. Therefore, monitoring the blood glucose values is very important. Since the non-invasive blood glucose estimation does not introduce the pain to the subjects and it can reduce the risk of the infection, the non-invasive blood glucose estimation techniques were recently developed. However, it is based on the photoplethysmograms (PPGs). Nevertheless, the PPGs are contaminated by the noise. Hence, performing the denoising is critical. To address this issue, this paper proposes a Slant transform based bit plane denoising method to improve the quality of the PPGs. First, the Slant transform is applied to the PPGs. Second, each Slant transform coefficient is represented using a finite number of bits. Third, some bits of the Slant transform coefficients are discarded. Fourth, the inverse Slant transform is applied to the processed PPGs. To perform the blood glucose estimation, the heart rates are extracted from the denoised PPGs and the heart rate variabilities are calculated. They are taken as the features. Then, the feature vectors and the corresponding invasive blood glucose values are used to train a random forest regression model. For the test PPGs, the non-invasive blood glucose values are estimated based on the established model. Compared to the states of the art methods, it is found that our proposed method can effectively improve the ability of estimation. In particular, the mean absolute relative difference (MARD) and the percentage of the data falling in the zone A of the Clarke error grid yielded by our proposed method reaches 15.34%, and 75.86%, respectively.