Analysis overview

There are three main ways we can analyse individual images:

1. Image classification - looking at the entire image and determining information about the subject
   1. In our case, this is typically used for example to identify if a plate is overgrown, has condensation or contamination or if a plate is in the image at all
2. Object detection
   1. Used to identify individual objects on the plate (bacterial or fungal colonies, seeds, insects, haloes)
   2. Various metrics are associated with each detected object including it’s position in the image, the size and the color values.
3. Object classification
   1. Used to identify what each individual object on the plate is (e.g. is it a seed, a shoot or a root or e.g. a yeast/bacterial colony vs a mold colony)

In the example below, each colony is identified through object detection and assigned an ID. Because there is a timeseries of images, the time when the object was first detected is accessible along with the size and color at each timepoint.

Each frame is analysed individually, but a time-series of images (a video) can also utilise our time tracking algorithms which allow for example the same colony growing on a plate to be tracked across time and thereby understand it’s growth rate or changes in morphology over time.

AI-based models run on each well in each plate and use machine learning models which are trained specifically to identify regions of interest on the plate such as a colony growing, a clearing zone forming, or even things like the size of a plantlet as it grows.

Annotations can be turned on in the user interface to understand exactly what a given model is picking up.

Real-time analysis running in the user interface with blue overlays indicating the analysis

The CFU detecting model looks at each frame of each plate during the timelapse as illustrated below:

Example of annotations during the timelapse of a single plate

Running at full throughput, each imaging device can image and analyze 15 petri dishes or 10 microtiter plates (up to 96-well) at a time. It is possible to run dozens of imaging devices in parallel, allowing for very high throughput automatic quantification.

Example job with 15 triplicate fungal colonies analysed by a CFU detector model

Plot of the dataset below from the 15-plate job example above

Plot of the dataset below from the 15-plate job example above

The area of each area identified is one of the most common phenotype metrics, but a variety of other data points are also available including various colorimetric outputs (mean RGB or HSV across the area) and the center coordinate for each area.

Analysis data can be exported in CSV format directly from the analysis export tab. 

Analysis export panel with options on analysis model, lighting mode and data format