infocus #69 March 2023 Using image processing to quantify biologically relevant information in flourescence microscopy images

10.22443/rms.inf.1.248

The aim of this summer project was to acquire images of fluorescently labelled tubulin in Schizosaccharomyces pombe cells and to produce an image processing pipeline adapted to quantifying the microtubules in these images.

One approach to detecting these linear structures was based on thresholding and skeletonisation, using plugins available in Fiji/ImageJ. This workflow yielded generally satisfactory results in the simplest case but was prone to creating artefacts and to loss of information. Crucially, this method could not cope with intersecting lines. As a more powerful alternative, the linear Hough transform was used to detect straight lines.

The implementation in Python was efficient and overcame some of the drawbacks of the previous method. However, it required significant post-processing to faithfully detect multiple lines in images. Unfortunately, it was only feasible to parametrise simpler instances of straight lines, and the method could not be extended to e.g., curved lines. Deep learning approaches could be an exciting front for future work on this problem.