Principal component-based image segmentation: a new approach to outline in vitro cell colonies

Identification, segmentation and counting of stained in vitro cell colonies play a vital part in biological assays. Automating these tasks by optical scanning of cell dishes and subsequent image processing is not trivial due to challenges with, e.g. background noise and contaminations. Here, we present a machine learning procedure to amend these issues by characterising, extracting and segmenting inquired cell colonies using principal component analysis, kk-means clustering and a modified watershed segmentation algorithm to automatically identify visible colonies. The proposed segmentation algorithm was tested on two data sets: a T-47D (proprietary) cell colony and a bacteria (open source) data set. High F-1 scores (similar to 0.90 for T-47D and >0.95 for bacterial images), along with low absolute percentage errors (similar to 11% for T-47D and <5% for bacterial images), underlined good agreement with ground truth data. Our approach outperformed a recent state-of-the-art method on both data sets, demonstrating the usefulness of the presented algorithm.

Automated summary

This article presents a machine learning procedure for automatic identification and segmentation of cell colonies, addressing challenges in image processing. The proposed algorithm achieves high accuracy and outperforms state-of-the-art methods on two tested datasets.