Deep-Learning-Based Blood Glucose Detection Device Using Acetone Exhaled Breath Sensing Features of α-Fe2O3-MWCNT Nanocomposites.

Abstract

Owing to the correlation between acetone in human's exhaled breath (EB) and blood glucose, the development of EB acetone gas-sensing devices is important for early diagnosis of diabetes diseases. In this article, a noninvasive blood glucose detection device through acetone sensing in EB, based on an α-Fe2O3-multiwalled carbon nanotube (MWCNT) nanocomposite, was successfully developed. Different amounts of α-Fe2O3 were added to the MWCNTs by a simple solution method. The optimized acetone gas sensor showed a response of 5.15 to 10 ppm acetone gas at 200 °C. Also, the fabricated sensor showed very good sensing properties even in an atmosphere with high relative humidity. Since the EB has high humidity, the proposed sensor is a promising device to exactly detect the amount of acetone in EB with high humidity. The sensor was powered by a 3200 mAh battery with the possibility of charging using mains electricity. To increase the reliability and calibration of the sensing device, a practical test was taken to detect acetone EB from 50 volunteers, and a deep learning algorithm (DLA) was used to detect the effect of various factors on the amount of acetone in each person's acetone EB. The proposed device with ±15 errors had almost 85% correct responses. Also, the proposed device had excellent response, short response time, good selectivity, and good repeatability, leading it to be a suitable candidate for noninvasive blood glucose sensing.