Z.S. Ballard, H-A Joung, A. Goncharov, J. Liang, K. Nugroho, J. Wu, D.K. Tseng, H. Teshome, L. Zhang, E.J. Horn, P.M. Arnaboldi, R.J. Dattwyler, O.B. Garner, D. Di Carlo
University of California, Los Angeles,
United States
Keywords: machine learning, point-of-care diagnostics, Lyme Disease
Summary:
We introduce a multiplexed vertical flow assay (xVFA) powered by deep learning for point-of-care (POC) diagnostic applications. The xVFA is comprised of different functional paper layers encapsulated by a disposable cassette that supports uniform vertical flow of buffer and serum through a multiplexed sensing membrane. After introduction of 20 µL of a given serum sample, gold nanoparticles, contained in a conjugate pad, are subsequently released upon wetting to produce a color signal via binding through a complementary detection antibody. After completion of the assay (~15 minutes), the xVFA cassette is opened and the sensing membrane is imaged with a custom-designed mobile phone reader, after which automated image processing calculates the colorimetric signal corresponding to each immunoreaction spot. Lastly, a trained deep neural network is used to rapidly infer a final result from the multiplexed immunoreaction information. This deep learning-based inference helps us mitigate inherent fabrication variations and operational tolerances borne out of the low cost paper-materials and the associated flow non-uniformities. To illustrate the power of this deep learning-based xVFA platform, two clinical POC use-cases are experimentally demonstrated, one for the early-stage Lyme Disease (LD) diagnostics, and the other for C-Reactive Protein (CRP) quantification used for heart disease risk-stratification. For the first application of LD testing, early diagnosis is critical for effective treatment of patients, however the current gold-standard testing methods suffer from high cost (>$400 per test), slow turn-around time (>24 hours), and poor sensitivity (