(291b) Integration of Electrochemical Sensing and Machine Learning to Detect Tuberculosis Via Methyl Nicotinate in Patient Breath

Integration of Electrochemical Sensing and Machine Learning to Detect Tuberculosis via Methyl Nicotinate in Patient Breath

Background and Motivation

Tuberculosis (TB) remains a pervasive global health issue, especially in low and middle-income countries, where access to rapid and accurate diagnostic methods is often limited. The World Health Organization's End TB Strategy emphasizes the need for innovative diagnostic tools that can operate at the point-of-care and are capable of sensitivity and specificity metrics that meet or exceed the threshold for a diagnostic test. Traditional diagnostic methods, such as sputum culture and molecular testing, are resource-intensive, time-consuming, and often not feasible in resource-limited settings. In addition, the rise of drug-resistant TB strains has heightened the urgency for early and accurate detection. Recent work by our lab, the Advanced Materials and Microdevices Lab at the University of Utah, has demonstrated the potential of non-invasive, rapid, and inexpensive electrochemical sensing methods designed to specifically detect volatile organic biomarkers (VOBs) in breath for TB diagnosis. Methyl-nicotinate (MN), a VOB associated with TB, presents a promising target for such diagnostics. Our study employs a novel integration of electrochemical sensing technology and XGBoost, a machine learning algorithm, to detect MN in patient breath samples, demonstrating the potential of this method in providing an accessible, inexpensive TB diagnostic.

Methods and Results

In our clinical demonstration, conducted in Kampala, Uganda, we analyzed breath samples from a cohort of 57 individuals, comprised of 42 microbiologically confirmed TB-positive patients and 15 TB-negative controls. The detection system utilized a copper(II) liquid metal salt solution in conjunction with square wave voltammetry tailored to detect MN with high accuracy. The implementation of an XGBoost machine learning model, with optimized hyperparameters and feature extraction, allowed for the analysis of complex patterns in electrochemical data, correlating them with TB status. This approach yielded an accuracy of 78%, with a sensitivity of 71% and specificity reaching 100%. These initial results support the model's potential in distinguishing TB-positive patients from negative controls based on MN concentration in breath samples.

Significance

The preliminary results from our study represent a significant step forward in the development of accessible, non-invasive, and specific VOB detection technology for TB diagnostics. By combining electrochemical analysis with machine learning, we demonstrate the feasibility of detecting TB through breath analysis, a method that holds potential for rapid, point-of-care testing. This approach not only aligns with the WHO’s goals for TB diagnosis in terms of sensitivity and specificity but also offers a scalable solution that could significantly reduce the diagnostic gap in TB-endemic regions. The success of this study paves the way for further research and development of breath-based diagnostics, with the ultimate aim of facilitating early detection and treatment initiation, thereby contributing to the global efforts to end TB.