Automated Cell Identification Methods for Stained Images using FIJI

Abstract

Accurate identification of stained cells in images is critical for many applications. One of the current methods is manual identification which is time consuming and has risk of human bias. Standard image processing methods consist of basic thresholding and water shedding. However, they struggle with segmenting cells when there are variations in cell morphology, large range of fluorescence intensity, and background artifacts. In addition to false positives and false negatives, errors include overcounting, undercounting, and inaccurate cell areas. In this project, we developed an automated cell counting approach in ImageJ (Fiji) that incorporates a series of optimized image processing steps that include background subtraction, edge detection, and adaptive thresholding. We also investigated using information from Hoechst nuclear staining to assist with segmentation of Calcein-stained images showing cell metabolism. In order to assess the performance of the various approaches, we manually identified cells in a set of challenging images and compared the results of the existing and proposed automated methods. Results show that the method improves throughput and also provides a robust tool for analysis of complex sample images.