Implementation of Machine Learning for Analysis of an on Demand Passive Sweat Cortisol Sensor

Journal Title
Journal ISSN
Volume Title

Cortisol is a steroid hormone produced by the adrenal glands for the purpose of regulating the body’s response to stress. Stress, as a physiological condition, can be caused by a wide variety of factors, such as mental exertion, diet, sleep, exercise, etc. For this reason, cortisol has the potential to serve as a biomarker for general health, as it relates to the everyday habits of patients. With the development of wearable technologies such as the smartwatch, increased attention has been focused on the development of noninvasive sensors for on demand testing that can integrated with wearable technologies. Current biosensing technologies for monitoring of chemical biomarkers such as cortisol depend on blood or salivary testing, which is invasive, costly, and time consuming. For this reason, the focus of this research is on the detection of cortisol through passive sweat, which contains many of the biomarkers present in blood at concentrations sufficient for detection. We have developed a noninvasive sensor on a flexible, nano porous substrate that has the capability to detect cortisol passively through sweat. The sensor data was then processed and input into a machine learning algorithm to analyze the rising and falling trend of cortisol concentration with time. The use of machine learning to analyze cortisol trends can be used to inform the wearer of rising or falling cortisol levels, which can enable them to make informed decisions about their health and lifestyle. Sensor response was measured by conducting Electrochemical Impedance Spectroscopy (EIS) assays of synthetic sweat dosed with concentrations of cortisol within the physiological range, from which the responses for low, medium, and high concentrations of cortisol were found to be significant. Similar assays were performed within the frequency region of maximum capacitance with dosing regimens ranging from high to low and low to high concentrations of cortisol, to simulate the rise and fall of cortisol levels of a human patient over a short period of time. The assay data was analyzed to find the rate of the change of the sensor response to a shift in cortisol concentration, which was then used to train a weighted KNN supervised machine learning algorithm to detect and classify increasing and decreasing cortisol concentrations in sweat. Algorithm accuracy was validated to be 100% by k means cross validation, showing that a passive, wearable sweat sensor can successfully be used with machine learning to detect rising and falling trends in cortisol concentration for on demand, noninvasive cortisol sensing.

Engineering, Biomedical