Manual and automated colony counters: selection and verification

A colony counter helps read visible colonies on a plate. It does not automatically identify microorganisms or prove the viability of every classified object from the image alone. Selection requires checking the whole workflow from plate to approved result.

A useful comparison covers the laboratory’s plates, including difficult ones. High correlation or short processing time does not establish that the result supports the QC decision. The tables and case below are original evaluation tools, not compendial criteria.

1. Define the reading task

Describe plate format and material, medium, any membrane, matrix, expected morphologies, observed densities and reading window. Include small or irregular colonies, colonies near the edge and plates without colonies. The scope must state when the system can count and when it must flag an uninterpretable condition.

Link counting to the use of the data: trending, process control or a result near a decision threshold. A few misclassified objects can have different consequences depending on the context. Microbiological recovery limitations are covered in bioburden and method suitability; this article assesses the reading system without changing culture conditions.

2. Compare manual, assisted and automated approaches

In manual reading, illumination, contrast, magnification, ergonomics and reader rules influence the result. An electronic tally can reduce recording errors without necessarily deciding which objects are colonies.

An automated system acquires images and applies classification rules or models. It may standardize reading and preserve a visual record, but still depends on acquisition and configuration. Young et al., 2018 showed that morphology and density affect comparisons between human and automated reading. This is not a guarantee transferable to every pharmaceutical plate.

3. Look for errors in challenging images

Do not crop the edge or remove a problematic area without a justified rule: real colonies may disappear too. Image quality and interpretability come before the displayed number.

4. Build a comparison that reveals differences

Prepare a representative set of plates and images spanning format, medium, density and visual difficulty. Justify sample size and repetitions against variability and the decision. Separate cases used to tune the system from those used to verify a fixed configuration; otherwise, the exercise may measure adjustment to the same examples.

Define a comparator, such as independent readings by qualified operators with a rule for resolving disagreements. Retain the initial readings: consensus is not uncertainty-free truth. Compare differences per plate, signed error, absolute error, repeatability and consequences for result classification. Assess low counts, dense cases and corrections separately.

Set criteria linked to use before the study, not a percentage selected after seeing the data. Percentage differences become unstable near zero and are undefined when the comparator is zero. Reprocessing the same image with a deterministic algorithm does not assess variability from fresh acquisition.

5. Control parameters and corrections

Manage reading area, thresholds, size filters and separation rules as a controlled configuration. A parameter should not change plate by plate to obtain an expected value. Define when manual correction is permitted, who may perform it and which outcomes require review.

Retain the original image, initial markers, edit, operator, reason and final result. In Heuser et al., 2023, visual review improved agreement but removed the time advantage in the configuration studied. Measure handling, correction, review and export as well as software processing time.

6. Qualify the system and protect records

Connect requirements, installation and operational checks, local performance and training. Supplier test images may check particular functions but do not replace verification on the laboratory’s plates. A counter is not validated simply because it uses an artificial intelligence model.

Identify camera, illumination, software, algorithm and versions. Verify plate-image-result association, access, relevant audit trails, transfers and record retrieval. A printout of the number may not preserve everything needed to reconstruct the analysis: assess original data, configuration and correction history using FDA data integrity principles.

7. Monitor routine use without hiding difficult cases

Simulated case, not an experiment: three plates yield the following values. The manual comparator is an agreed reading for the example, not a certified truth.

Both averages are approximately 33.3 colonies per plate and the mean signed difference is zero. Yet two plates differ by ten colonies; mean absolute error is about 6.7. Cancellation between overestimation and underestimation hides the problem. Three cases illustrate the risk; they are not a validation plan.

For routine use, define functional checks, reader competency, monitoring of corrections and actions for unexplained differences. New plates, matrices, lighting or updates require impact assessment and proportionate testing before use. Return to the microbiology laboratory hub to connect reading, method and quality system.

Sources and access limitations

• WHO, TRS 961, Annex 2, 2011: good practices for equipment, methods and quality in pharmaceutical microbiology laboratories.
• Young et al., 2018, PMID 29341168; Heuser et al., 2023, PMID 37395656: public abstracts consulted. Studies of specific systems and plate sets, not universal pharmaceutical QC criteria.
• FDA, Data Integrity and Compliance With Drug CGMP, December 2018: original data, context and review. Sources checked on 30 September 2026.