Non-invasive blood glucose detection system based on conservation of energy method

• Published in: Physiological measurement 2017-02, Vol.38 (2), p.325-342
• Main Authors: Zhang, Yang, Zhu, Jian-ming, Liang, Yong-bo, Chen, Hong-bo, Yin, Shi-min, Chen, Zhen-cheng
• Format: Article
• Language: English
• Subjects: Adolescent; Aged; back propagation neural network; Blood Glucose Self-Monitoring - instrumentation; Blood Glucose Self-Monitoring - methods; conservation of energy law; decision tree; Decision Trees; Diabetes Mellitus - blood; Diabetes Mellitus - metabolism; Energy Metabolism; Female; glucose monitoring; Humans; Male; Middle Aged; Neural Networks (Computer); non-invasive; Photoplethysmography; Young Adult
• ISSN: 0967-3334; 1361-6579
• DOI: 10.1088/1361-6579/aa50cf

The most common method used for minimizing the occurrence of diabetes complications is frequent glucose testing to adjust the insulin dose. However, using blood glucose (BG) meters presents a risk of infection. It is of great importance to develop non-invasive BG detection techniques. To realize high-accuracy, low-cost and continuous glucose monitoring, we have developed a non-invasive BG detection system using a mixed signal processor 430 (MSP430) microcontroller. This method is based on the combination of the conservation-of-energy method with a sensor integration module, which collects physiological parameters, such as the blood oxygen saturation (SPO2), blood flow velocity and heart rate. New methods to detect the basal metabolic rate (BMR) and BV are proposed, which combine the human body heat balance and characteristic signals of photoplethysmography as well dual elastic chambers theory. Four hundred clinical trials on real-time non-invasive BG monitoring under suitable experiment conditions were performed on different individuals, including diabetic patients, senior citizens and healthy adults. A multisensory information fusion model was applied to process these samples. The algorithm (we defined it as DCBPN algorithm) applied in the model combines a decision tree and back propagation neural network, which classifies the physiological and environmental parameters into three categories, and then establishes a corresponding prediction model for the three categories. The DCBPN algorithm provides an accuracy of 88.53% in predicting the BG of new samples. Thus, this system demonstrates a great potential to reliably detect BG values in a non-invasive setting.