Exhale-Dx™: A non-invasive, real-time breath analysis system using deep learning for asthma diagnosis.

Background: Asthma presents significant diagnostic and therapeutic challenges, impacting millions and posing a substantial burden on healthcare systems, particularly in the United Kingdom, where it afflicts roughly 5.4 million individuals. Severe asthma, incurring over 50% of total expenditures, tends to lead to frequent exacerbations and preventable emergency admissions. Traditional diagnostic methods, primarily based on clinical history, can result in delays and misdiagnoses, culpable for over 1,200 deaths annually, 90% of which are considered preventable with timely intervention.

Objective: To address this issue, we developed Exhale-Dx™, a point-of-care breath test platform that provides a non-invasive, user-friendly solution for asthma diagnosis and monitoring. Exhale-Dx™ captures volatile organic compounds (VOCs) in exhaled breath, reflecting real-time metabolic and inflammatory markers of lung function. By analyzing these personalized breath signatures, clinicians and patients can detect exacerbations up to three days in advance, thus facilitating early and targeted interventions to reduce emergency care utilization. The system integrates capnographic waveforms, asthma control scores, and clinical lung function data, offering a comprehensive diagnostic profile.

Methods: Using Exhale-Dx™ data, we developed the Asthma Diagnostic Enhanced Neural Architecture (ADENA), an advanced deep neural network that leverages VOC biomarkers and lung function data to enhance diagnostic precision.

Results: ADENA achieved exceptional performance, delivering 98.7% accuracy, an F1 score of 0.98, and a low mean squared error of 0.065. The deconvolution analysis further confirmed the model's ability to detect significant physiological differences between asthmatic and non-asthmatic profiles.

Conclusion: Our findings showed that VOC analysis combined with advanced neural networks could accurately distinguish asthmatic profiles, highlighting their potential for early, non-invasive interventions in respiratory health diagnostics.