Viral infections pose a significant global health burden, exemplified by influenza, HIV, HPV, and the COVID-19 pandemic. Accurate and affordable virus quantification is crucial for clinical diagnosis, vaccine development, and antiviral agent production. The plaque assay, developed in 1952, remains the gold standard for quantifying replication-competent lytic virions. However, traditional plaque assays involve manual counting, are time-consuming (2–14 days incubation), and prone to human error. While alternative methods exist, such as immunofluorescence focal forming assays, PCR, and ELISA-based assays, they often lack the cost-effectiveness and infectivity assessment capabilities of the plaque assay. Recent advancements in quantitative phase imaging (QPI), holography, and deep learning offer a potential solution to overcome these limitations. QPI provides non-invasive, label-free visualization and quantification of transparent biological specimens, and its image quality can be enhanced using neural networks for tasks like phase retrieval, noise reduction, auto-focusing, and resolution improvement. Numerous deep learning methods have been successfully applied to microorganism detection and identification using QPI. This paper presents a cost-effective, compact, label-free, live plaque assay for significantly faster and more accurate PFU quantification than traditional methods, eliminating the need for staining.

The existing literature highlights the limitations of traditional plaque assays, including the lengthy incubation times (2-14 days) and the inherent subjectivity of manual plaque counting, leading to potential inaccuracies and inter-observer variability. While numerous attempts have been made to improve the efficiency and accuracy of plaque assays, many of these methods still rely on fluorescence markers or specialized culture plates, increasing costs and complexity. The lack of a truly rapid, automated, and cost-effective plaque assay remains a significant gap in virology research and clinical applications. This research builds upon the growing body of work demonstrating the capabilities of quantitative phase imaging (QPI), lens-free holography, and deep learning for various biological applications, including automated cell counting and classification. These technologies offer the potential to create a significantly faster and more accurate method for plaque assay quantification while mitigating the shortcomings of existing techniques.

A compact lens-free holographic imaging prototype was constructed at a cost of <$880 (excluding a standard laptop). This system rapidly scans a six-well plate hourly (~0.32 gigapixels per well scan), using the reconstructed phase images for PFU detection based on spatiotemporal changes. A neural-network-based classifier converts the phase images into PFU probability maps, revealing PFU locations and sizes. The system was tested on Vero E6 cell plates infected with VSV, HSV-1, and EMCV. For VSV, a machine learning-based coarse PFU localization algorithm was used during training to accelerate dataset generation and delineate false positives. A total of 357 true-positive and 1169 negative holographic videos were collected and augmented, resulting in 2,594 positive and 3,028 negative videos for training the PFU classifier. The network was blindly tested on 30 wells, converting holographic phase images into PFU probability maps (~7.5 min per well). A 0.5 probability threshold was used to obtain final PFU detection results. The system's performance was compared against traditional plaque assays (48 h incubation and crystal violet staining) and an Agilent BioTek Cytation 5 device with Gen5 software. For HSV-1 and EMCV, transfer learning was employed using the trained VSV network as a base model, achieving significant incubation time reduction without introducing false positives. High-concentration virus samples were tested to assess the device's ability to handle a broader dynamic range. The study also explored the relationship between virus concentration and infected cell area percentage, demonstrating the system's ability to provide earlier PFU readouts based on infected area quantification. Detailed descriptions of cell propagation, virus propagation, agarose preparation, well plate preparation, crystal violet staining, lens-free imaging setup, image pre-processing, coarse PFU localization algorithm, network training dataset generation, network architecture and training schedule, image post-processing, automated PFU counting algorithm, and automated PFU counting settings for the BioTek Cytation 5 are provided in the Methods section.

The developed stain-free device achieved a detection rate of >90% at 20 h of incubation for VSV, significantly faster than the 48 h required for traditional assays, with 100% specificity. This represents a substantial reduction in incubation time. Compared to the Agilent BioTek Cytation 5 system, the holographic method achieved a similar detection rate (93.7% vs 94.3%) but with 0% false discovery rate and 28 h earlier detection. The system demonstrated strong generalization ability, achieving an 89% detection rate on 12-well plates without retraining. Transfer learning successfully adapted the system to HSV-1 and EMCV, reducing incubation times by 48 h and 20 h, respectively, with >90% detection rate and 0% false positives. The device effectively handled high virus concentrations, providing reliable PFU quantification even in cases of severe spatial overlapping, whereas the traditional assay was limited to lower concentrations. The system showed resilience to artifacts from cell viability problems. A strong correlation was observed between virus concentration and infected cell area percentage, allowing for early PFU concentration estimates. For samples with higher virus concentrations, the infected cell area percentage reached >1% in ≤10 h, providing even earlier PFU concentration readouts. The lens-free holographic system provided a high throughput of ~0.32 gigapixels in <30 s per well, enabling rapid scans of entire six-well plates.

This study successfully demonstrates a cost-effective, automated, and significantly faster method for viral plaque quantification. The combination of lens-free holography and deep learning allows for label-free, high-throughput imaging and automated PFU detection, addressing the key limitations of traditional plaque assays. The system's ability to detect PFUs in their early stages of growth, without staining, represents a major advancement. The system's robustness to artifacts and its adaptability to different viruses further enhances its practical utility. The quantitative relationship between infected area and virus concentration opens avenues for even faster assays. The results strongly suggest the potential for wide-scale adoption of this technique across diverse virology applications.

This research presents a novel stain-free, rapid, and quantitative viral plaque assay leveraging lens-free holography and deep learning. The system significantly reduces incubation time, eliminates staining, provides high accuracy and throughput, and is adaptable to various viruses. This technology holds substantial promise for accelerating virology research, vaccine development, and clinical diagnostics. Future work could explore parallel imaging, improved scanning stages, and multi-wavelength phase recovery to further enhance the system's performance.

While the system demonstrates high accuracy and speed, potential limitations include the need for initial training data generation for each virus type, although transfer learning can mitigate this for related viruses. The accuracy of the PFU quantification relies on the quality of the trained neural network, which could be affected by the size and variability of the training dataset. The current system is optimized for specific cell types and virus types; further optimization might be required for other applications.