Estimation of Glycated Hemoglobin Using Multiwavelength Photoplethysmography Signal

Diabetes mellitus is a metabolic disorder, and pro-longed exposure to sugar molecules in the blood stream results in hemoglobin glycation. It is measured in percentage of glycated hemoglobin $(\mathbf{HbA}_{1\mathrm{c}})$ cells in all hemoglobin cells. A suspected person may have symptoms such as abrupt weight loss, increased thirst and hunger, frequent urination, etc. The conventional method requires invasive blood testing methods performed in specific laboratories. The work proposes a non-invasive solution to measure $\%\mathbf{HbA}_{1\mathrm{c}}$ using Photoplethysmography (PPG) signal. A PPG sensor records these signals at two distinct wavelengths, i.e., green and infrared (IR). After initial preprocessing and feature extraction, a random forest regressor estimates the $\%\mathbf{HbA}_{1\mathrm{c}}$. The study is validated with recording of 9 subjects. The estimation of the results shows a high degree of correlation with 0.907 and 0.952 pearson's correlation coefficient $\mathbf{r}$ values for the reflective PPG signal. These calculated values are compared with the true $\%\mathbf{HbA}_{1\mathrm{c}}$ values. The results show a close approximation of the calculated values to the true values. Also, the error analysis is performed in terms of mean squared and root mean squared error.