Volatile Organic Compound Profiler for Use in Future Glucose Monitoring

Volatile organic compounds (VOCs) have been shown to appear in the exhaled breath of an individual albeit in miniscule quantities. What researchers have found however, is that the concentrations of specific VOCs found in our breath can reveal useful information regarding one’s physiological state. A potential use for this could be the future research and development of noninvasive glucose monitoring systems. The issue however, is that it’s difficult to accurately measure the concentrations of VOCs found in a breath sample. To aid with the measurement and profiling of VOCs, machine learning algorithms are used with Tedlar bags. This has the disadvantage of being expensive, time consuming, and overall, inefficient as it requires a very specific set of expensive tools in a lab setting. There needs to be a way to train algorithms for use in VOC profiling in a way that's relatively inexpensive, easy to use, and efficient in that it is not dependent on a patient being in a typical lab setting. Our proposed solution is a small handheld device capable of accepting a trained machine learning algorithm, displaying results on a user friendly graphical user interface (GUI), and saving data on an external memory card, all enclosed in a 3D printed case. More specifically our device will include an algorithm that is trained to be able to differentiate the VOC profiles found in Purell and Germ-X brands of hand sanitizers in order to correctly identify the brand used. The mobility and ease of use of our system will contribute to the development of sophisticated machine learning algorithms for use in VOC profiling, which in of itself will allow for further developments of noninvasive medical solutions for glucose monitoring. To accomplish our goal we’re using an STM32 development board, a TFT touchscreen, three Figaro VOC sensors, and a machine learning algorithm that we are developing based on Keras. Currently we have successfully gathered preliminary sensor data with our VOC sensors, demonstrating how the profiles of two brands of hand sanitizer, Purell and Germ-X , are distinct enough to differentiate. We’ve also created an early iteration of our machine learning algorithm that is capable of successfully outputting the classification of the data with 80% accuracy. Additionally, we were able to create a simple GUI on our touchscreen with limited functionality capable of interacting with an Arduino. Combined, these results are significant because they illustrate the major theories and assumptions of our project will successfully work as anticipated.