Hypoglycemia detection: multiple regression-based combinational neural logic approach

• Published in: Soft computing (Berlin, Germany) Germany), 2017, Vol.21 (2), p.543-553
• Main Authors: Ling, Sai Ho, San, Phyo Phyo, Lam, Hak Keung, Nguyen, Hung T.
• Format: Article
• Language: English
• Subjects: Alarm systems; Artificial Intelligence; Boundary conditions; Case studies; Computational Intelligence; Control; Design; Diabetes; Diabetes mellitus; Electrocardiography; Engineering; Feature selection; Genetic algorithms; Glucose monitoring; Heart rate; Hybrid systems; Hypoglycemia; Insulin; Logic; Mathematical Logic and Foundations; Mechatronics; Methodologies and Application; Monitoring; Monitoring systems; Neural networks; Parameters; Particle swarm optimization; Physiological effects; Regression; Robotics; Support vector machines; Telemedicine
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-015-1809-z

Hypoglycemia is a common and serious side effect of type 1 diabetes. We measure physiological parameters continuously to provide detection of hypoglycemic episodes in type 1 diabetes mellitus patients using a multiple regression-based combinational neural logic approach. In this work, a neural logic network with multiple regression is applied to the development of non-invasive hypoglycemia monitoring system. It is an alarm system which measures the physiological parameters of electrocardiogram signal (heart rate and corrected QT interval) and determine the onset of hypoglycemia by the use of proposed hybrid neural logic approach. In this clinical application, a combinational neural logic network with multiple regression is systematically designed to hypoglycemia detection based on the characteristic of this application. To optimize the parameter of the hybrid combinational neural logic system, hybrid particle swarm optimization with wavelet mutation is applied to tuned the parameters of the system. To illustrate the effectiveness of the proposed method, hypoglycemia monitoring system which will be practically analyzed using real data sets collected from 15 children ( 14.6 ± 1.5 years) with type 1 diabetes at the Department of Health, Government of Western Australia. With the use of proposed method, the best testing sensitivity of 79.07 % and specificity of 53.64 % were obtained.