Petri CounterBack to the workspace

RESEARCH ABSTRACT

Methods prototype

Petri Counter: browser-based, human-assisted enumeration of yeast colonies with reviewable separation of touching growth

Max Rogers

Counting engine v0.1.0

Abstract

Background. Enumeration of yeast colonies from plate photographs becomes ambiguous when adjacent colonies touch or image artifacts resemble growth. We developed Petri Counter to combine inspectable automatic center detection with manual correction and preservation of review annotations.

Methods. A deterministic pipeline designed for dark colonies on pale agar runs locally in a browser without a trained model or GPU. Images are resampled to a maximum long-axis dimension of 1,400 pixels. Color-based dish localization defines an editable elliptical region, with an adjustable inset. Background correction and contrast thresholding identify candidate foreground regions. Local maxima of a smoothed distance transform propose colony centers within each connected region; separation and saddle-depth criteria suppress duplicate peaks. Size, intensity, and shape heuristics reject implausible candidates, and regions crossing the counting boundary are excluded. Users can inspect detections, adjust the dish outline, and add or remove centers. Saved submissions retain the original photograph, automatic and corrected markers, settings, self-reported reviewer names, and review scope.

Results. Synthetic regression checks recovered the specified counts for isolated circular objects, chains of two through six touching objects, a branched six-object cluster, and a tested unequal-size touching configuration. Additional checks covered empty scenes, a smooth illumination gradient, simulated dark-blue ink, and boundary exclusion. These checks establish behavior on constructed cases, not biological counting accuracy. Ten supplied plate screenshots serve as exploratory examples; expert-adjudicated reference counts and center annotations were unavailable for this report. Consequently, real-plate count error, detection precision and recall, split and merge error rates, and time savings remain unmeasured.

Conclusions. Petri Counter provides an inspectable prototype for assisted colony enumeration and collection of candidate reference annotations. Validation requires independent expert review and held-out plates spanning colony density, morphology, and acquisition conditions. Fully fused growth may not retain enough visual information to resolve individual colonies. Saved feedback does not automatically retrain the counter.

Keywords Yeast colonies; colony enumeration; distance transform; touching colonies; image analysis; human review

TOWARD BIOLOGICAL VALIDATION

The next study

01

Establish reference annotations

Define inclusion rules for small colonies, dish edges, and unresolved confluence. Collect independent expert center annotations, adjudicate disagreements, and document organism, medium, incubation, and imaging conditions.

02

Evaluate on held-out plates

Keep plates and acquisition sessions separate from those used for tuning or training. Report plate-level count error, center-matching precision and recall, and split and merge errors, with uncertainty estimates and results by density and image quality.

03

Keep review provenance

Preserve independent submissions and unresolved regions. Review labels require adjudication before use as reference data. Compare automatic counts and assisted review separately; measure review time if claiming an efficiency benefit.

Related work

These publications provide context for colony-image analysis and dataset evaluation. Petri Counter has not been benchmarked against their systems or datasets.

  1. Majchrowska S, et al. AGAR a microbial colony dataset for deep learning detection. arXiv:2108.01234 (2021). A reference resource for annotated plate imagery.
  2. Signoroni A, et al. Hierarchical AI enables global interpretation of culture plates in the era of digital microbiology. Nature Communications 14, 6874 (2023). A separate approach to automated culture-plate interpretation.