oa Noninvasive Diabetes Monitoring with Electronic Nose

Background

Diabetes is a chronic illness in which the body cannot manage the levels of sugar, and it affects an estimated 387 million people worldwide. According to World Health Organization it is the 7th leading cause of death in the world. Frequent monitoring (3 to 10 times) of blood glucose is important for diabetes patients to prevent serious health problems, but existing commercial products of blood glucose measurement often cause discomfort and pain. In these products, a lancet is used for pricking the skin to obtain a blood drop and then this blood drop is placed on the test strip for quantification of blood glucose level with a meter. This inconvenient, painful and invasive technology is the main obstacle to achieve high quality health monitoring. Providing a pain-free and noninvasive diabetes monitoring solution at an affordable cost is the major challenge to replace the existing solutions.

Objective

Based on the recent experimental findings about breath acetone as a potential correlated biomarker for changes in blood glucose, we introduce an electronic nose based breath analyzer solution for identification/ quantification of exhaled acetone and hence, blood glucose level.

Method

The experimental setup used to acquire the signatures of the acetone with the electronic nose, containing an array of four low-cost gas sensors, is shown in Fig. 1. Mass flow controllers (MFCs) are used to control the concentration of acetone in the gas chamber from 0.25 to 2.5 parts per million (ppm), which corresponds to high point for breath acetone concentration. The sensor array, placed in a gas chamber, is periodically exposed to air for 750 seconds and to acetone for 500 seconds to extract meaningful information or feature vector at each concentration in the target range. The feature vector is formed by taking the ratio (a.k.a. sensitivity) of the change in each sensor resistance during the gas exposure stage with its baseline resistance (resistance at the end of air exposure). The resultant feature vector is further used for acetone identification/quantification. For acetone identification, we form sensitivity codes by arranging their sensitivities in an ascending order and introduce hardware friendly rank-order classifier. After its identification, support vector regression is used for its quantification by utilizing feature vectors.

Results

Acetone data at 10 different concentrations in the target range of 0.25 to 2.5 ppm is acquired to evaluate the performance of our approach. Figure 2 shows the typical response of the four sensors in the array corresponding to air (from 0 to 750 seconds) and acetone (from 751 to 1250 seconds). On testing with resultant feature vectors from the experimental data, rank-order classifier identifies acetone signature with 100% accuracy, and then support vector regression (SVR) quantifies acetone with a mean square error of 0.0059. Predicted concentration with SVR against true concentration in the target range is shown in Fig. 3.

Conclusion

We have introduced a breath-based acetone monitoring solution which will not only be eagerly acceptable by the diabetes patients due to its noninvasive and pain-free nature but it will also facilitate high quality health monitoring by acquiring a large number of breath samples at an affordable cost. Test results of the proposed electronic nose system illustrate good response of the sensors to acetone induced by breadth.