EAGER: COVID-19 Real-time Detection via Hyperspectral Analysis of Sweat Metabolite Biometrics

The project seeks to address the ongoing challenge pertaining to the Covid-19 outbreak and the need for prompt diagnosis in a reality where the Test Kits are scarcely available, expensive, labs-based and slow. The analysis of biometric metabolites has the potential to be clinically applicable in monitoring the health of individuals based on particular biomarker combinations.

This exploratory study will evaluate the sensitivity and specificity of sweat metabolite biometrics for detecting COVID19 infection in human subjects with and without symptoms. The proposed methodology involves a radically different testing approach while engaging novel interdisciplinary perspectives: medical technology, artificial intelligence and machine learning are combined to prevent and diagnose an infectious serious disease. This will reduce the cognitive burden on humans by promoting human-machine teaming for improved overall detection performance. The project seeks to develop tools that will enable real-time and accurate screening of COVID-19 applicable on a large-scale population, aiding the community to face further spread of it. Through monitoring of the disease biomarkers in sweat, the proposed method has also the advantage of being non-invasive. The methods, theory, and data resulting from this proposal will impact the scientific community in several positive ways and will be made publicly available through an appropriate website. Findings and results achieved through this project will enhance the engineering curricula, including image processing, biometrics and machine learning. Advances produced with this project will be disseminated by the investigators through publications and web seminars. The proposed research will create a new hyperspectral imaging methodology to acquire sweat metabolites specifically impacted by COVID-19 to be processed through pattern recognition strategies. Hyperspectral imaging is a powerful tool for non-destructive analysis, enabling real-time monitoring of spatially resolved spectral information of materials.

This proposal seeks to design, extract and evaluate features from hyperspectral data cubes, stacked images across pre-defined wavelengths, using a compact hyperspectral imager without involving chemical methods. A novel dataset of hyperspectral fingerprint images will be developed during this research study. Advanced image processing techniques will be used to extract rich signals and machine learning for training pattern classifiers to distinguish between diseased and non-diseased subjects. In addition, this project seeks to build a bridge between recent advances in chemistry and image processing, creating a novel effective representation of COVID-19 profiles based on discriminative biomarkers quantified in terms of concentrations in the hyperspectral domain. The concentrations of the biochemical content in human sweat have been measured using reagent kits and instruments such as spectrophotometers. With this research, relevant spatial information will be integrated with corresponding spectral signature to enable the diagnosis through artificial intelligence. The proposed COVID-19 detector will be assessed using standard performance metrics of machine learning algorithms and compared to tampon-based testing methods.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.