Non-invasive blood glucose monitoring for diabetics by means of breath signal analysis

Much attention has been focused on the non-invasive blood glucose monitoring for diabetics. It has been reported that diabetics' breath includes acetone with abnormal concentrations and the concentrations rise gradually with patients' blood glucose values. Therefore, the acetone in human breath can be used to monitor the development of diabetes. This paper investigates the potential of breath signals analysis as a way for blood glucose monitoring. We employ a specially designed chemical sensor system to collect and analyze breath samples of diabetic patients. Blood glucose values provided by blood test are collected simultaneously to evaluate the prediction results. To obtain an effective classification results, we apply a novel regression technique, SVOR, to classify the diabetes samples into four ordinal groups marked with 'well controlled', 'somewhat controlled', 'poorly controlled', and 'not controlled', respectively. The experimental results show that the accuracy to classify the diabetes samples can be up to 68.66%. The current prediction correct rates are not quite high, but the results are promising because it provides a possibility of non-invasive blood glucose measurement and monitoring.