Method for training machine learning model for blood glucose measurement based on Raman spectroscopy
Blood glucose level measurement includes a light source configured to irradiate light to a subject; a monochrome part configured to separate wavelength components of the light that is reflected and scattered from the subject; a light receiver configured to receive the light transmitted via the monochrome part and to generate electrical signals based on the received light; and a processor configured to extract information on the blood glucose level of the subject based on a frequency shift of the light due to the Raman effect.
1. A method comprising:
causing a light source to irradiate light to a subject;
receiving, from a light receiver, electrical signals representative of a Raman spectrum of the light, which is reflected and scattered from the subject, and wavelength components of which are separated by using a monochrome part;
obtaining areas of (a) a peak at about 1450 cm −1 , (b) a peak at about 1660 cm −1 , and (c) a peak at about 1125 cm −1 in the Raman spectrum;
receiving a measured blood glucose level of the subject; and
updating a machine learning model based on a correlation of the measured blood glucose level of the subject with the areas obtained using the Raman spectrum,
wherein, for (c) the peak at about 1125 cm −1 , the area is obtained by summing areas in three ranges (c-i) between 1089 cm −1 and 1160 cm −1 , (c-ii) between 1115 cm −1 and 1140 cm −1 , and (c-iii) between 1120 cm −1 and 1130 cm −1 .
2. The method of claim 1 , wherein the measured blood glucose level of the subject is measured using at least one of finger blood sampling, vein blood sampling, or continuous CGM.
3. The method of claim 1 , wherein the machine learning model utilizes partial least squares (PLS), support vector machine (SVM), autoencoder, ResNet, or any combination thereof.
4. The method of claim 2 , wherein the machine learning model comprises a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), a deep Q-network, a generative adversarial network (GAN), or any combination thereof.
5. The method of claim 1 , wherein, for (a) the peak at about 1450 cm −1 , the area is obtained in a range between −1415 cm −1 and −1480 cm −1 .
6. The method of claim 1 , wherein, for (b) the peak at about 1660 cm −1 , the area is obtained in a range between −1630 cm −1 to −1685 cm −1 .