DiabeticSense: A Non-Invasive, Multi-Sensor, IoT-Based Pre-Diagnostic System for Diabetes Detection Using Breath

Abstract

Diabetes mellitus is a widespread chronic metabolic disorder that requires regular blood glucose level surveillance. Current invasive techniques, such as finger-prick tests, often result in discomfort, leading to infrequent monitoring and potential health complications. The primary objective of this study was to design a novel, portable, non-invasive system for diabetes detection using breath samples, named DiabeticSense, an affordable digital health device for early detection, to encourage immediate intervention. The device employed electrochemical sensors to assess volatile organic compounds in breath samples, whose concentrations differed between diabetic and non-diabetic individuals. The system merged vital signs with sensor voltages obtained by processing breath sample data to predict diabetic conditions. Our research used clinical breath samples from 100 patients at a nationally recognized hospital to form the dataset. Data were then processed using a gradient boosting classifier model, and the performance was cross-validated. The proposed system attained a promising accuracy of 86.6%, indicating an improvement of 20.72% over an existing regression technique. The developed device introduces a non-invasive, cost-effective, and user-friendly solution for preliminary diabetes detection. This has the potential to increase patient adherence to regular monitoring.

Noninvasive Diabetes Detection through Human Breath Using TinyML-Powered E-Nose

“…This comprehensive approach would bolster generalization and performance, culminating in a more precise and robust embedded system for breath analysis across diverse patient cohorts. Ultimately, such advancements would provide invaluable support for medical professionals in their diagnostic and monitoring endeavors [ 35 , 55 , 59 ].…”

“…This approach meets all the requirements of cutting-edge AI computing and enables the efficient processing and classification of data quality [ 57 , 58 ], independent of resource-consuming cloud services or a computer for classifying new VOC-collected data [ 59 ]. In human breath analysis, the integration of TinyML and sensors as a noninvasive technique has been effective in predicting respiratory diseases such as chronic obstructive pulmonary disease (COPD) [ 60 ].…”