Blood glucose level prediction for diabetic patients using intelligent techniques

Eskaf, K

Abstract

Diabetes mellitus is one of the most common chronic diseases. The number of cases of
diabetes in the world is likely to increase more than two fold in the next 30 years; from
115 million in 2000 to 284 million in 2030. This work is concerned with helping diabetic
patients to manage themselves by trying to predict their blood glucose level (BGL) after 30
minutes on the basis of the current levels in order that they can administer insulin. This
will enable the diabetic patient to continue living a normal day life activities as much as is
possible.
In order to achieve this objective, three techniques were developed and evaluated: a
Numerical Analysis algorithm, an Artificial Neural Network (ANN), and a Genetic
Algorithm (GA). In the case of the ANN and the GA, the variation in Blood Glucose
Levels was modelled as a Mass Spring Damper, treating the food intake as a bolus
injection of glucose, and thus the impulse force F (f), and the effects of exercise and
hypoglycaemic medication were represented by the damping factor, p. The values of F, f$
and the differences in BGL every 5 minutes were used as knowledge features in the
training and prediction phases for the ANN and GA.
Data was derived for a virtual diabetic patient from a web-based educational simulation
package for glucose-insulin levels in human body using the AIDA software. The Dexcom
SEVEN System was used to capture the BGLs of two diabetic patients and a normal
person for 24 hours with a sampling frequency of 5 minutes. The two databases were used
in all prediction algorithms.
Newton's Interpolatory Divided Difference (Numerical Analysis) algorithm was used to
predict the future BGLs and found to be able to predict the level after 5 minutes from the
current value of BGL with a RMSE less than 0.5 mmol/1. Unfortunately, the RMSE
increased above 2.5 mmol/1 when trying to predict 15 or 20 minutes ahead. The ANN
using Feed Forward Back Propagation was able to predict the BGL after 30 minutes with a
RMSE between 0.49 mmol/1 to 1.8 mmol/1, while the GA was found to predict the BGL
30minutes ahead with a RMSE between 0.15 mmol/1 to 0.42 mmol/1.
It is concluded that the GA provided the best technique for prediction in this application.

Citation

Eskaf, K. Blood glucose level prediction for diabetic patients using intelligent techniques. (Thesis). Salford : University of Salford

• Thesis Type: Thesis
• Deposit Date: Oct 3, 2012
• Award Date: Jan 1, 2011