End-to-end plaque counting and virus titration from laboratory plate images with deep learning
This paper introduces Titra, an end-to-end deep learning workflow that automates the entire virus titration process—from well detection and plaque segmentation to PFU/mL estimation—demonstrating strong agreement with manual annotations across diverse viral species and plate formats.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
In the quiet corners of virology laboratories, scientists rely on a time-honored method to measure how much infectious virus is present in a sample. They grow cells in a dish, introduce the virus, and wait for the virus to kill the cells in small, visible circles called plaques. Each circle represents a single infectious particle that started a chain reaction of destruction. By counting these circles, researchers can calculate the virus concentration, a critical number for understanding how a virus spreads or how well a vaccine works. For decades, this counting has been done by human eyes, a slow and tedious process that varies from person to person and is prone to fatigue. The challenge has always been to find a way to automate this counting without losing the precision that human experts provide, especially when the circles are tiny, crowded, or blurry.
A team of researchers has now introduced a new digital workflow called Titra that takes a photograph of a laboratory plate and automatically counts these viral circles, calculating the final virus concentration in a single step. Unlike previous tools that only handled one part of the job, such as finding the circles or just counting them, this system manages the entire process from the raw image to the final biological number. The researchers tested their system on images of three different viruses: Mayaro virus, Coxsackievirus B3, and vaccinia virus. They used plates with different numbers of wells and images taken under varying lighting conditions, including photos snapped with a standard smartphone camera in a regular lab. The system first identifies the individual wells in the plate, then locates every viral circle within them, and finally separates circles that are touching or overlapping to ensure an accurate count.
The results showed that the automated counts matched the work of human experts with remarkable consistency. For the Mayaro and Coxsackievirus samples, the computer's counts correlated almost perfectly with manual counts, and the final calculated virus concentrations were nearly identical to those derived by hand. Even for the vaccinia virus, which produced much smaller and more crowded circles that are notoriously difficult to count, the system performed strongly, though it did struggle slightly when the circles were so dense they merged into a single mass. The researchers found that while the computer sometimes made small mistakes, such as spotting a tiny speck as a circle where none existed, these errors were usually minor and did not significantly alter the final result. In fact, the system's performance was often as reliable as the agreement between different human experts, who themselves sometimes disagreed on how to count the most ambiguous or crowded circles.
What makes this approach particularly useful is that it does not require expensive, specialized cameras or perfect laboratory lighting. The system was designed to work with standard photographs, making it accessible for routine use in labs that do not have high-end imaging equipment. The researchers also built the tool into a web-based platform where a human expert can quickly review the computer's work. If the system is unsure about a specific circle or if the circles are too crowded to separate clearly, the expert can step in to correct the count. This creates a partnership where the machine handles the heavy lifting of scanning thousands of images, while the human provides oversight for the difficult cases. The study suggests that this hybrid approach can drastically reduce the time and effort required for virus testing while maintaining the high standards of accuracy needed for scientific research.
The researchers were careful to note that while the system works well across different viruses and plate types, it is not a perfect solution for every possible scenario. When the viral circles are extremely dense and overlapping, the system, like the human eye, finds it harder to distinguish individual units. However, the study demonstrated that even in these challenging conditions, the final virus concentration estimates remained reliable. The work also highlighted that measuring the success of such a system requires looking beyond just how well it draws the outline of a circle; the true test is whether the final number it produces matches the biological reality. By focusing on the end goal of accurate virus measurement rather than just the technical details of image processing, the team created a tool that bridges the gap between complex artificial intelligence and practical laboratory needs.
This new workflow represents a shift in how routine biological measurements are performed, moving away from purely manual labor toward a system that combines computational power with human judgment. The researchers plan to make their software and the data they collected available to the public, allowing other scientists to use and improve the tool. As the technology evolves, it could potentially be adapted for other types of biological assays, expanding its utility beyond just counting viral circles. For now, it stands as a proven method for making the tedious task of virus titration faster, more consistent, and more accessible to laboratories around the world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.