CellVision: A deep learning based image analysis platform to accelerate immuno-plaque assay data processing for...

Accurately detecting patterns of interest across a large number of images presents a significant challenge in data analysis for high-throughput analytical assays. In many cases, manual image review to some degree is necessary, deviating from a fully automated workflow desired in high-throughput platforms. CellVision, a deep learning-based workflow, was created using recent developments in AI-powered image analysis to enable truly automated and accurate counting of viral plaques. This Python-based workflow uses image processing techniques and a convolutional neural network to identify viral plaques, separate fused plaques, and differentiate plaques from other objects that appear in the same image. By integrating pattern recognition and classification methods, CellVision enables fully automated data analysis for the high-throughput µPlaque assay. Its design accommodates a variety of atypical cases found in real-world images, facilitating the identification of diverse patterns. The high accuracy achieved by CellVision eliminates the need for a manual, visual review of the images, substantially reducing time to data release. CellVision has been fully integrated into the µPlaque assay workflow and has been used to support an investigational, live-attenuated, quadrivalent dengue vaccine in development by Merck & Co., Inc., Rahway, NJ, USA. The accuracy of plaque counting results from CellVision was demonstrated against a diverse dataset verified by multiple analysts, outperforming an alternative commercial tool. CellVision's successful integration into the assay workflow and its potential for generalization to other viral plaque assays signify a new standard in applying AI-powered methods to antiviral vaccine discovery and development.