Graph based method for cell segmentation and detection in live-cell fluorescence microscope imaging. (January 2022)

• Record Type: Journal Article
• Authors: Hajdowska, Katarzyna; Student, Sebastian; Borys, Damian
• Is Part Of: Biomedical signal processing and control. Volume 71(2022)Part A
• Publication Date: 2022-01
• Subjects: Microscope image processing -- Cell segmentation -- Hough transform -- Watershed segmentation -- Graph cut segmentation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103071
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We are proposing a new approach to live-cell fluorescence image segmentation – an algorithm GRABaCELL. Proposed method uses Graph Cut, Watershed segmentation and Hough Circular Transform. We also propose a new accuracy metric to determine the quality of segmentation. Proposed method presents better results in terms of Dice and proposed metric than other methods compared in this article. The method has been validated on multiple datasets. Abstract: Live-cell fluorescence image segmentation is an essential step in many studies, including in drug research and other contexts where keeping cells alive is crucial. Several segmentation algorithms and programs have been previously proposed; however, they do not work sufficiently well on top-down pictures with overlapping cells. Our proposed algorithm, called GRABaCELL, utilizes Graph Cut, Watershed segmentation and Hough Circular Transform to improve automatic segmentation and counting living cells. We also introduce a modified accuracy metric to determine the quality of segmentation in terms of the number of cells detected in the image. The GRABaCELL method results are vastly better in visual assessment, by both Dice index and modified accuracy metric, than all other compared methods maintaining not only a high value of these indices but also a relatively small spread.