Computer Vision-aided CRISPR Diagnostics for the Detection of COVID-19

Abstract

Surveillance testing is a key strategy to control the spread of COVID-19. Unlike the gold standard testing method, quantitative reverse transcription polymerase chain reaction (RT-qPCR), CRISPR diagnostics have recently become a more appealing alternative as they are proven to be faster, simpler, and more affordable. However, the current CRISPR diagnostic readouts are typically non-quantitative, making them error-prone and lacking crucial information of viral load. To further improve the CRISPR diagnostic method, we have developed a custom computer vision algorithm that works in complement to common transilluminators to process fluorescence images of the diagnostic samples, quantify their fluorescence signals, and assign the test results. Our analysis showed that the quantified fluorescence intensity was directly correlated to the sample viral load, useful information for transmissibility and disease severity. Verified through laboratory and clinical samples, our algorithm accurately discriminated the samples with the viral RNA as low as 6.25 copies/uL, and correctly classified nasopharyngeal swab (NP swab) samples with 100% accuracy. Our work serves as a potential technique to improve the accuracy of CRISPR diagnostics of COVID-19 and promote rapid testing vital to the containment of the ongoing pandemic.