End-to-end plaque counting and virus titration from laboratory plate images with deep learning
- 1Arionkoder
Abstract
Plaque assays are the gold standard for quantifying infectious virus, yet plaque enumeration is still routinely performed manually, making virus titration labor-intensive, subjective, and difficult to standardize across analysts and laboratories. Existing automated methods primarily address individual tasks, such as plaque segmentation or counting, but do not provide an integrated workflow from plate images to biological quantification. In this paper we present Titra, an end-to-end workflow for automated cytopathic effect (CPE)-based virus titration from standard photographs of plaque assay plates. The workflow combines automatic well detection, plaque segmentation, post-processing for instance separation, plaque counting, and plaque-forming units per millilitre (PFU/mL) estimation within a single web-based platform that also enables experiment management and expert review. The approach was evaluated using images from three viral species (Mayaro virus, Coxsackievirus B3, and vaccinia virus), two plate formats (6- and 12-well), and heterogeneous image acquisition conditions, including both a newly curated dataset and the public VACVPlaque dataset. Automated plaque counts showed strong agreement with manual annotations (Pearson correlation coefficients of 0.98 for MAYV/CVB3 and 0.88 for VACV), while PFU/mL estimates closely matched manual calculations for the MAYV/CVB3 dataset (Pearson r=0.975). Comparative experiments against U-Net, StarDist, HSD-WBR, and PyPlaque demonstrated competitive segmentation and counting performance, with the proposed approach achieving the highest Dice and mAP on the public VACVPlaque dataset.
Related
- A membrane-impermeant nucleic acid dye converts bacteriophage plaque assays into a machine-readable format for automated counting
- AgentCount: An Agentic Workflow for Traceable Colony Counting and Insight Generation from Agar Plate Images
- Proteomic machine-learning signatures associated with carotid plaque: A population-based hypothesis-generating study
- Multi‐omics integration and machine learning identify an ACKR1+ endothelial cell‐based lactate‐immune signature associated with carotid plaque vulnerability and progression
- Deep Learning Based Cell Segmentation of 3D Confocal Images of Wood for Enzymatic Hydrolysis Applications
BibTeX
@misc{moris2026endtoend,
title = {{End-to-end plaque counting and virus titration from laboratory plate images with deep learning}},
author = {E. Moris and A. Costable and S. Rey and I. Ferreiro and J. Hurtado and M. Villagran and L. L. Luciano and A. E. Vazquez and J. Ramos and I. Monteiro and M. V. de Santiago and P. Moreno and G. Moratorio and J. I. Orlando},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.04.749375},
url = {https://www.biorxiv.org/content/10.64898/2026.09.04.749375v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - End-to-end plaque counting and virus titration from laboratory plate images with deep learning
AU - E. Moris
AU - A. Costable
AU - S. Rey
AU - I. Ferreiro
AU - J. Hurtado
AU - M. Villagran
AU - L. L. Luciano
AU - A. E. Vazquez
AU - J. Ramos
AU - I. Monteiro
AU - M. V. de Santiago
AU - P. Moreno
AU - G. Moratorio
AU - J. I. Orlando
PY - 2026
DA - 2026/09/04
PB - bioRxiv
DO - 10.64898/2026.09.04.749375
UR - https://www.biorxiv.org/content/10.64898/2026.09.04.749375v1
AB - Plaque assays are the gold standard for quantifying infectious virus, yet plaque enumeration is still routinely performed manually, making virus titration labor-intensive, subjective, and difficult to standardize across analysts and laboratories. Existing automated methods primarily address individual tasks, such as plaque segmentation or counting, but do not provide an integrated workflow from plate images to biological quantification. In this paper we present Titra, an end-to-end workflow for automated cytopathic effect (CPE)-based virus titration from standard photographs of plaque assay plates. The workflow combines automatic well detection, plaque segmentation, post-processing for instance separation, plaque counting, and plaque-forming units per millilitre (PFU/mL) estimation within a single web-based platform that also enables experiment management and expert review. The approach was evaluated using images from three viral species (Mayaro virus, Coxsackievirus B3, and vaccinia virus), two plate formats (6- and 12-well), and heterogeneous image acquisition conditions, including both a newly curated dataset and the public VACVPlaque dataset. Automated plaque counts showed strong agreement with manual annotations (Pearson correlation coefficients of 0.98 for MAYV/CVB3 and 0.88 for VACV), while PFU/mL estimates closely matched manual calculations for the MAYV/CVB3 dataset (Pearson r=0.975). Comparative experiments against U-Net, StarDist, HSD-WBR, and PyPlaque demonstrated competitive segmentation and counting performance, with the proposed approach achieving the highest Dice and mAP on the public VACVPlaque dataset.
ER -