CFU counting

The "CFU counting" analysis pipeline is a robust general-purpose AI-based model for counting any sort of colony on plates, either based on an end-point image or throughout a time-series.

On top of counting the total number of colonies, it has the ability to differentiate colonies, either through various phenotypes (e.g. color, size, time of appearance, growth rate), or by directly estimating what type of colony is present (e.g. differentiating yeast from mold or a suspect pathogen colony from a non-pathogen).

The CFU counting pipeline can handle a wide variety of methods, including:

• Surface plating

• Pour plating

• Environmental/air

• Membrane filter dishes

Examples

Example of an environmental monitoring plate with AI-based counting and detection of colonies of differing morphology

Reporting formats

The most common output format for CFU counting will be a simple CSV file with the plate position and enumeration result. The result will be given as a count. If the plate is too numerous to count, by default the result will be labelled "TNTC". Typically, anything above 300 colonies on a 90mm petri dish would be labelled TNTC, but it is possible for the model to count up to 1000+ distinct colonies on a plate (resolution down to 0.1

It is possible to input sample names (for example barcodes) directly into the Reshape platform through the user interface or through automated sample tracking.

For certain applications, it's desirable to be able to add comments/flag certain edge cases. This could for example be the presence of various debris, dried out plates or similar. This is also possible as part of the pipeline output, and is usually reported with a "comment" column:

Often, a differentiated count is needed. There are several typical distinctions depending on the specific method (such as yeast/mold differentiation) - typically, this would be reported as separate columns as in the example below:

Time to result reduction

With the timelapsing capabilities of the Reshape platform, additional capabilities are added to the traditional end-point method.

Every single experiment is digitalized, and therefore, it is very easy to produce a dataset documenting the exact distribution of incubation time required to reach a CFU result. This can be used to cut down the incubation time, particularly in R&D environments. For certain methods like fungal spore enumeration, the time reduction can be up to 70%, for example taking a 7-day method down to just 2 days. Because the imaging and analysis happens in real-time, it's also possible to follow along in any given experiment from the office (both the quantification but also seeing the plates themselves).

Example of a time-series count, with the orange curve representing the count at each time-point, allowing for example an accurate documentation of the required time-to-result. The blue curve represents the cumulative area of colonies on the plate

Accuracy

Validation is typically part of any implementation to document the accuracy of the platform versus a human reference count. The exact accuracy is dependent on the specific methods, but typically, the accuracy exceeds the repeatability between human operators.