Electronic Nose using Quartz Tuning Fork Sensors and Machine Learning for Detection of Ketoacidosis

Ketoacidosis is a metabolic state with high concentration of ketone bodies in blood and urine, which occurs when the body shifts from carbohydrate to fat metabolism for energy production. This poses significant health risks to individuals if not detected and treated promptly. Traditional diagnostic methods can be invasive and time-consuming, highlighting the requirement for alternate diagnostic methods. This study investigates the potential of breath analysis to diagnose ketoacidosis by detecting acetone, a biomarker in breath. An array of six quartz tuning fork based sensors was employed to detect acetone, isoprene, and acetaldehyde. The sensor response was collected by varying the analyte concentrations from 8 ppm to 100 ppm. The responses from four were used to predict the analyte concentration using machine learning. A feed-forward neural network regressor was employed for prediction. The regressor was able to predict individual gas concentration with a mean squared error of 3.6 ppm and an R-squared value of 0.98. The work demonstrates that the combination of an array of quartz tuning fork sensors along with a machine learning model can detect acetone at lower concentrations and is a non-invasive diagnostic tool for early detection of ketoacidosis, potentially improving patient outcomes through more accessible and timely monitoring.