MCount: An automated colony counting tool for high-throughput microbiology

Accurate colony counting is crucial for assessing microbial growth in
high-throughput workflows. However, existing automated counting solutions
struggle with the issue of merged colonies, a common occurrence in
high-throughput plating. To overcome this limitation, we propose MCount,
the only known solution that incorporates both contour information and
regional algorithms for colony counting. By optimizing the pairing of
contours with regional candidate circles, MCount can accurately infer the
number of merged colonies. We evaluate MCount on a precisely labeled
Escherichia coli dataset of 960 images (15,847 segments) and achieve an
average error rate of 3.99%, significantly outperforming existing
published solutions such as NICE (16.54%), AutoCellSeg (33.54%), and
OpenCFU (50.31%). MCount is user friendly as it only requires two
hyperparameters. To further facilitate deployment in scenarios with
limited labeled data, we propose statistical methods for selecting the
hyperparameters using few labeled or even unlabeled datapoints, all of
which guarantee consistently low error rates. MCount presents a promising
solution for accurate and efficient colony counting in application
workflows requiring high throughput, particularly in cases with merged
colonies.