The COVID-19 pandemic, caused by SARS-CoV-2, necessitates rapid and accessible diagnostic tools. While RT-qPCR is the gold standard, its limitations (cost, invasiveness, specialized labs) hinder widespread use, especially in developing countries. This study explores the use of an electronic nose as a complementary diagnostic tool. Electronic noses analyze volatile organic compounds (VOCs) in exhaled breath, offering a noninvasive, rapid, and potentially low-cost alternative. Previous research has shown the potential of electronic noses to detect disease-specific VOC patterns, but further development and validation are necessary to establish their clinical utility. This paper focuses on developing and validating a novel electronic nose for COVID-19 screening.

Existing literature highlights the limitations of RT-qPCR for widespread COVID-19 testing, particularly in resource-limited settings. Several studies have explored the use of electronic noses for disease diagnostics, with promising results in identifying disease-specific VOC patterns in exhaled breath. These studies demonstrate the potential of electronic nose technology to provide a non-invasive, rapid, and relatively inexpensive alternative to conventional methods. However, the specificity and sensitivity of these methods vary across studies due to differences in sensor arrays, feature extraction techniques, and machine learning algorithms. This study aimed to improve on these existing methods by creating a new electronic nose with optimized parameters and more advanced data analysis techniques.

The researchers developed a portable electronic nose, GeNose C19, consisting of two main units: a sensing unit and a breath sampling unit. The sensing unit contains an array of 10 metal oxide semiconductor gas sensors housed in a 3D-printed chamber. The breath sampling unit includes a HEPA filter to remove particulate matter and water molecules from the exhaled breath before reaching the sensors. A disposable medical-grade polyvinyl chloride bag collects the breath sample. The collected breath is then introduced into the sensor chamber via a PTFE tube. The sensors' responses are recorded, and feature extraction is performed using four parameters: maximum, median, standard deviation, and variance. Four machine learning algorithms (Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Stacked Multilayer Perceptron (MLP), and Deep Neural Network (DNN)) were employed for pattern recognition and classification of breath samples. The study used a prospective case-control cohort design, enrolling 83 subjects (43 COVID-19 positive, 40 negative) confirmed by RT-qPCR. 70% of samples were used for training, and 30% for testing. The performance of GeNose C19 was assessed using sensitivity, specificity, accuracy, and the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. Additional analysis was conducted to evaluate the impact of several factors including sensor cross-sensitivity to temperature and humidity, and VOC composition analysis using Gas Chromatography-Mass Spectrometry (GC-MS).

GeNose C19 demonstrated high performance in differentiating between COVID-19 positive and negative breath samples. The Deep Neural Network (DNN) model yielded the best results with sensitivity and specificity levels of 95.5% (95% CI: 92.7–97.3%) and 95.7% (95% CI: 92.7–97.5%), respectively, and an AUC of 95.6% (95% CI: 93.7–97.1%). Across all four machine learning algorithms, the system exhibited accuracy levels ranging from 88% to 96%. The study also investigated the cross-sensitivity of the sensors to temperature and humidity, finding that these factors could be controlled and minimized through preconditioning protocols. GC-MS analysis revealed no significant differences in VOC concentration between positive and negative groups, suggesting that the discriminatory power of GeNose C19 relies on the pattern of VOCs rather than their concentration. However, some VOCs (e.g., acetone, ethyl butyrate) showed different trends between groups, aligning with findings from other studies. The DNN model consistently outperformed other algorithms in terms of stability and performance, showing minimal differences between training and testing data.

The high accuracy, sensitivity, and specificity of GeNose C19, particularly using the DNN model, demonstrate its potential as a rapid and noninvasive screening tool for COVID-19. The non-invasive nature, rapid analysis time (approximately 3 minutes), and low cost make it a suitable alternative to RT-qPCR, especially in resource-constrained settings. The finding that the discriminatory power relies on VOC patterns rather than individual concentrations is significant, suggesting the potential for accurate classification even with variations in individual VOC levels. The use of a HEPA filter effectively removes viral particles from the breath sample, ensuring the safety of the device and personnel. The study acknowledges limitations regarding the potential influence of confounding factors such as comorbidities, diet, and environmental VOCs. Further research with larger and more diverse populations is necessary to validate GeNose C19's performance in real-world settings and explore its potential as a diagnostic tool.

GeNose C19 demonstrates significant promise as a rapid, non-invasive, and potentially low-cost screening tool for COVID-19. The high accuracy and ease of use make it a valuable potential addition to existing diagnostic methods. Future research should focus on validating GeNose C19's performance in larger, more diverse populations, including those with comorbidities and different ethnic backgrounds, and exploring its ability to differentiate COVID-19 from other respiratory infections.

The study's limitations include its open-label design, the relatively small sample size, and the potential influence of confounding factors such as diet, comorbidities, and environmental VOCs. The reliance on RT-qPCR as the gold standard also introduces potential for false positives and negatives. Further research with larger and more diverse populations, employing a double-blinded randomized controlled trial, is necessary to fully validate the clinical utility of GeNose C19.