AI-Empowered Resonating Microstructures for Non-Invasive Real-Time Breath VOCs Analysis and Tracking
The identification and quantification of volatile organic compounds (VOCs) in exhaled breath serve as valuable indicators for disease diagnosis and prognosis. Existing sensing methods, however, face challenges in terms of achieving sensitivity, selectivity, and portability within a same device. This research proposes a novel approach that utilizes dielectric microstructure arrays to enhance image-based analysis coupled with deep learning techniques to overcome the above-mentioned limitations. Unlike traditional optical methods, our approach employs microstructures with resonating wavelengths coinciding with the IR absorption wavelengths of target VOCs, ensuring superior sensitivity and selectivity. The design incorporates pre-separation and time-of-flight mechanisms for efficient gas separation during travel. Multiple stations with dielectric microstructure array-based sensing platform can be established and by capturing intensity profiles at various wavelengths using a camera, and a deep learning model can extract features for analysis. Ground truth validation can be obtained through gas chromatography. This innovative approach can be a breakthrough in the simultaneous detection and quantification of multiple VOCs in breath with high selectivity and sensitivity utilizing a simple pre-separation method, compact optical sensing unit and augmented the efficiency by deep learning technique presenting a promising avenue for various healthcare applications. This visionary methodology not only promises to redefine the landscape of healthcare applications but stands as a beacon of innovation, offering a transformative and unparalleled solution to the complexities of VOC analysis in breath diagnostics.
Keywords: Breath biomarker, gas sensing, micro-structure, deep-learning, health-care
呼氣中揮發性有機化合物（VOC）的識別和定量可以作為疾病診斷和預後（prognosis）的指標。然而，現有的感測方法無法同時實現高靈敏度、高選擇性和便攜性方面的權衡面。故本研究提出了一種新穎的方法：利用介電微結構陣列以增強基於影像的分析，並結合深度學習以克服上述限制。與傳統的光學方法不同，此方法採用諧振波長與目標 VOC 的紅外光吸收波長一致的微結構，以確保高靈敏度與高選擇性。該設計結合了預分離與time‐of‐flight機制，可實現有效的氣體分離。感測平台的多個量測站將由介電微結構陣列組成，透過以紅外光相機捕捉不同波長的強度分佈，並結合深度學習模型以進行特徵萃取與分析，後透過氣相層析法驗證。本研究提出的方法可以同時辨識和定量呼吸中的多種VOC，具有高選擇性和靈敏度，利用簡單的預分離方法、小體積的光學感測單元並結合深度學習技術以提高效率，在各種醫療保健應用上極具潛力，為複雜的呼吸診斷的VOC分析提供變革性的視角與解法。
關鍵字：呼吸生物標誌物，氣體感測，微結構，深度學習，醫療保健
