Ask Rapid: Understanding Algorithms

Answering your questions about automated image analysis and the Growth Direct® System. 

In a recent post, we outlined Avoiding 4-Eyes Uncertainty and Regulatory Risk for microbiology laboratories that still rely on manual observation and recording of plate count results. Bypassing the need for slow, error-prone human eyes is a goal of automated QC technologies such as the Growth Direct® System, but you might be wondering: how does a machine judge what it’s looking at? 

The answer can be found in the Growth Direct® System’s image analysis software, which finds and counts fluorescent objects as they grow in size and intensity over time. Here’s how it works.  

Inside our system’s imaging station are multiple light emitting diodes (LEDs) that expose incubated system cassettes to blue light, causing the flavins in any living cell to radiate green fluorescent light. This natural autofluorescence is in turn detected by a large charge-coupled device (CCD) panel, similar to those used in astronomy photography and other high-quality imaging applications. Long before the naked eye can detect a colony forming unit (CFU), our imaging hardware can digitally “picture” colonies with as few as 100 cells and capture the x-y coordinates needed to track that growth over time. 

Separating discrete CFUs from all the “noise” in such hyper-detailed imaging is just one challenge. For the purpose of QC microbiology, our software must satisfy stringent requirements for counting consistency, reduction of incubation time, data capture, and data integrity; to do so, it applies proprietary detection and enumeration algorithms that we developed here at Rapid Micro Biosystems.  

What is an algorithm, really? 

All sorts of news and entertainment stories these days talk about algorithms without defining the term, so it’s easy to be a little fuzzy on the concept. Broadly speaking, an algorithm can be considered a step-by-step procedure for solving a problem or accomplishing a task while satisfying a defined set of rules. When you follow a recipe for making a cake or baking a pizza, you’re essentially carrying out an algorithm.   

In the case of the Growth Direct® System, our algorithms are carefully defined, automated procedures which have been proven to yield accurate results across a wide range of microorganism genus/colony shapes and sizes, even in the face of potentially interfering materials such as dyes and particulates.  

For example, an algorithm sequence to distinguish fluorescent objects from the surrounding membrane and media captured in a digital image would include these steps: 

• Clusters of neighboring CCD pixels whose intensity significantly exceed the local background are classified as objects​ 
• Changes to key morphological parameters of the objects, such as intensity, and size, are monitored by the algorithm over time to identify the objects as either growing micro-colonies or non-growing debris 

Algorithms, then, allow us to standardize and automate many of the same judgments made by experienced microbiologists when they look at a test plate. This ensures that the Growth Direct® System delivers results that meet the conditions for an automated system of traditional microbial plate counting specified by USP <1223>, “allow[ing] colonies on solid media to be read more quickly, with substantially less incubation time than is possible using only the unaided eye.”  

Have more questions? Contact us today to discuss the Growth Direct® System and how it can help automate your QC Micro operations.