Non-invasive Glucose Level Estimation: A Comparison of Regression Models Using the MFCC as Feature Extractor

• Published in: Pattern Recognition pp. 206 - 215
• Main Authors: Francisco-García, Victor, Guzmán-Guzmán, Iris P., Salgado-Rivera, Rodolfo, Alonso-Silverio, Gustavo A., Alarcón-Paredes, Antonio
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2019
• Series: Lecture Notes in Computer Science
• Subjects: Mel frequency cepstral coefficients; MFCC; Non-invasive glucose measuring; Optical sensing
• ISBN: 9783030210762; 3030210766
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-21077-9_19

The present study comprises a performance comparison on well-known regression algorithms for estimating the blood glucose concentration from non-invasively acquired signals. These signals were obtained measuring the light energy transmittance of a laser-beam source through the fingertip by means of an embedded light dependent resistor (LDR) microcontroller system. Signals were processed by computing the Mel frequency cepstral coefficients (MFCC) to perform the feature extraction. The glucose concentration in blood was measured by a commercial glucometer in order to evaluate the performance of five well-known regression models. The experimental results revealed comparable values of mean absolute error (MAE) and Clarke grid analysis. The best performance was obtained by the support vector regression with a mean absolute error of 9.45 mg/dl. However, this study serves as a starting point and alludes to the potential application of non-invasive systems in the glucose level estimation. Future experiments measuring the glucose concentration with laboratory standard tests should be conducted, and a model implementation in an embedded device for their use is also mandatory.