Poster R17 - Rapid and Stain-free Quantification of Viral Plaque Enabled by Deep Learning and Holography

Introduction: To address the long-standing global threat of viral infections, a variety of virus detection and quantification techniques have been developed, significantly advancing the fields of vaccine and antiviral drug development. Among these techniques, the viral plaque assay stands out as the gold standard method owing to its unique ability to cost-effectively assess virus infectivity by observing the formation of plaques resulting from virus replication on a cell monolayer. However, traditional viral plaque assays present several challenges. First, they are time-consuming, typically requiring 2-14 days of incubation for plaque-forming units (PFUs) to grow to sizes visible to the naked eye. Additionally, chemical staining is necessary to introduce the color contrast of PFU samples for visual counting, which is reagent-consuming and involves laborious procedures. Furthermore, manual counting of PFUs is both tedious and prone to human errors. To address these challenges, we developed a rapid, accurate, and automated PFU detection system that can substantially reduce the detection time and completely eliminate the need for chemical staining and manual counting processes.

Materials and

Methods: PFU samples in our study were prepared according to the standard viral plaque assay protocol. We validated our method using Vero E6 cells infected with three types of viruses: vesicular stomatitis virus (VSV), herpes simplex virus type-1 (HSV-1), and encephalomyocarditis virus (EMCV). The VSV samples were first imaged using a custom-built holographic imaging device (Fig. 1(a)) for 20 hours (72 hours for HSV-1 and 60 hours for EMCV) with an imaging time interval of 1 hour (2 hours for HSV-1 and 1 hour for EMCV). Subsequently, the samples underwent further incubation for up to 48 hours (120 hours for HSV-1 and 72 hours for EMCV), after which they were fixed and stained to obtain the ground truth images (Fig. 1(c)), which were used for comparison. At each imaging/detection time point, the reconstructed holographic phase images of test plates were transferred into probability maps using a trained deep neural network-based PFU classifier (see Fig. 1(b)), which can distinguish true PFUs from other non-PFU artifacts by analyzing their unique spatiotemporal features. Each pixel on the probability maps indicates the probability of the local region centered at this pixel having a PFU or not. Finally, a threshold of 0.5 was applied to these probability maps to generate binary detection masks, which clearly indicate the sizes and locations of the PFUs.

Results, Conclusions, and Discussions: As shown in Fig. 1(d, f, h), the blind testing and detection results from our system showed high concordance with their corresponding stained ground truths, but with a significantly shorter detection time. Specifically, when tested on five 6-well plates (comprising a total of 335 VSV PFUs and five negative control wells), our label-free system achieved an average VSV PFU detection rate of 90.3% at 17 hours of incubation (Fig. 1(e)), saving >24 hours compared to the 48-h stained ground truth. In addition, our method successfully generalized to new types of viruses, HSV-1 and EMCV, through transfer learning. When blindly tested on two 6-well plates (totaling 214 HSV-1 PFUs and two negative control wells), it achieved a 90.4% HSV-1 detection rate at 72 hours (Fig. 1(g)), reducing the incubation time by 48 hours compared to 120 hours required by the traditional HSV-1 plaque assay. Similarly, a 90.8% EMCV detection rate was achieved at 52 hours of incubation (Fig. 1(i)) on two 6-well plates (totaling 249 EMCV PFUs and two negative control wells), saving 20 hours relative to the 72-hour traditional EMCV plaque assay. Furthermore, our system was able to efficiently handle samples with virus concentrations approximately 10 times higher than standard PFU assays by identifying individual PFUs during their early growth stages before cluster formation. In summary, our system offers a rapid, cost-effective, automated, and stain-free viral plaque assay technique. It provides reliable PFU counts while significantly reducing time, labor, and reagent use, making it a valuable tool for both laboratory and field applications.

Acknowledgements (Optional):