Artificial intelligence (AI)-based digital health diagnostics has the potential to leverage the development of sensitive portable detection technologies for timely management of emerging pathogens. In our work, we aimed to capitalize on the advances in deep learning to develop a rapid (<50 mins), inexpensive (<$1), and accurate (100% with 62 patients) diagnostic system for use at the point-of-care. Our system was developed with a focus on rapid reconfiguration to different virus targets, to aid in the long-term epidemic preparedness strategies. We used our technology originally designed to detect HIV, Hepatitis B and C, and Zika infections and reconfigured it to detect COVID-19 using a few example cases. The updated AI-system was then validated with 62 patient samples imaged with a smartphone where the system performed with 100% accuracy in identifying patients with and without the virus. We are now deploying our assay onto an automated platform for easy-of-use, which performs the assay on the press of a button.
OBJECTIVE: The goal of this study is to develop sensitive portable detection technologies adaptive to emerging pathogens. We interfaced adversarial neural networks (ADVN)-based image processing to a simple non-enzymatic viral-detection-assay, to detect clinically-relevant ranges intact SARS-CoV-2 in nasal swab samples rapidly.
MATERIALS AND METHODS: Firstly, we developed and standardized a protocol that allowed detection of intact-viruses in a microfluidic chip, through readable optical signals (gas-bubbles) using smartphone-captured images. The dataset library (17,573 images) that consists of labeled- dataset for 4 different pathogens (HIV, HBV, HCV and, ZIKV) and unlabeled-dataset with images from simulated and synthetically generated samples was used in the development of ADVN. A small seed set of SARS-CoV-2 samples was used in reconfiguring this algorithm to detect COVID-19. Finally, the accuracy of this ADVN-model to classify negative and positive samples was evaluated using a separate dataset of nasal swab sample (n = 62) images.
CONCLUSION: The generalizability of the developed system was confirmed by rapid reconfiguration to detect SARS-CoV-2 antigens in nasal swab samples (n = 62) with 100% accuracy at a detection threshold of 1000 copies/mL. The assay was incorporated onto an automated platform which is currently being validated. Such a system can contribute to epidemic preparedness strategies.