Integration of electronic nose technology with spirometry: validation of a new approach for exhaled breath analysis

Introduction

Diagnostic tests are vital in modern respiratory medicine, aimed at guiding clinical management to improve patient outcomes. While methods like spirometry are routine, molecular diagnostics are less accessible at the point of care.

Exhaled breath contains thousands of volatile organic compounds (VOCs) from both systemic and local metabolic processes. Traditional breath analysis often uses gas chromatography with mass spectrometry (GCMS) to identify specific VOCs. In contrast, electronic noses (eNoses) detect patterns of gas mixtures without identifying individual compounds, using cross-reactive sensors.

Research in respiratory diseases shows that eNose technology can distinguish conditions like lung cancer, asthma, COPD, and infections with accuracy similar to standard tests. The SpiroNose, developed in collaboration with Amsterdam UMC and Comon-Invent BV, integrates eNose technology with spirometry, allowing for real-time, online breath analysis. This approach eliminates errors from indirect sampling and streamlines diagnosis at the point of care.

The study aimed to validate the SpiroNose’s accuracy and reliability in clinical settings, focusing on technical validation, signal processing, and diagnostic accuracy for routine use.

Methods

Subjects and Technical Validation: The study aimed to validate the technical performance of the SpiroNose®. It included 60 healthy adults for technical validation and 144 subjects for clinical validation, divided into four groups: asthma, COPD, lung cancer, and healthy controls. Patients were recruited from clinical sites in Amsterdam (AMC) and Enschede (MST), while controls were recruited via advertisements.

Design and Methodology: The validation included assessing the SpiroNose’s integration with the pneumotachograph and its robustness against variables like breath volume, airflow, humidity, and ambient substances. Clinical validation followed a cross-sectional case-control design during routine lung function testing, without dietary or medication restrictions. Measurements were performed in duplicate, with real-time data transmission to an online server for analysis.

Measurements: The SpiroNose® used five sensor arrays (four MOS sensors each) to analyze exhaled breath: three arrays for VOC analysis in breath and two for ambient VOC correction. The sensor stability and reproducibility were confirmed using a standard test gas before each session.

Data Analysis: The primary analysis focused on comparing exhaled breathprints among patients with asthma, COPD, lung cancer, and healthy controls. Additionally, breathprints from asthma patients at different sites (Amsterdam University Medical Centers vs. Medisch Spectrum Twente) were compared to assess the suitability of the methods for multi-center studies. Data were processed using Matlab and SPSS, with corrections made for ambient VOCs. The data were normalized and reduced via principal component analysis (PCA), followed by categorical classification. The pre-processed data from 12 sensors were reduced to four principal components using PCA. These components underwent ANOVA to identify those that best distinguished between groups. The discriminative components were used in linear discriminant analysis (LDA) to classify cases. Cross-validation (leave-one-out method) determined the model’s accuracy. ROC curves were constructed, and the area under the curve (AUC) was calculated. Reproducibility was assessed using Cohen’s kappa analysis.

Results

The clinical validation phase demonstrated the diagnostic power of the SpiroNose in differentiating between various respiratory diseases. The device accurately distinguished between asthma, COPD, and lung cancer, showing strong diagnostic performance. The accuracy for differentiating asthma from healthy controls reached 87%, with an AUC of 0.94. When distinguishing COPD from controls, the accuracy was 78%, with an AUC of 0.80. For lung cancer, the SpiroNose achieved an impressive 88% accuracy and an AUC of 0.95. Furthermore, the SpiroNose was able to differentiate between COPD and lung cancer with an accuracy of 87%, even when excluding lung cancer patients with co-existing COPD. The consistency of these results across different study sites further supports the robustness and generalizability of the technology.