Machine learning for enumeration of cell colony forming units

Abstract

As one of the most widely used assays in biological research, an enumeration of the bacterial cell colonies is an important but time-consuming and labor-intensive process. To speed up the colony counting, a machine learning method is presented for counting the colony forming units (CFUs), which is referred to as CFUCounter. This cell-counting program processes digital images and segments bacterial colonies. The algorithm combines unsupervised machine learning, iterative adaptive thresholding, and local-minima-based watershed segmentation to enable an accurate and robust cell counting. Compared to a manual counting method, CFUCounter supports color-based CFU classification, allows plates containing heterologous colonies to be counted individually, and demonstrates overall performance (slope 0.996, SD 0.013, 95%CI: 0.97–1.02, p value < 1e-11, r = 0.999) indistinguishable from the gold standard of point-and-click counting. This CFUCounter application is open-source and easy to use as a unique addition to the arsenal of colony-counting tools.