A Unet-based method for Cell Segmentation Challenge

Lanfeng Zhong

Keywords: Cell segmentation

TL;DR: A stardist-based method for cell instance segmentation

Abstract:

Automatic detection and segmentation of cells and nuclei in microscopy images is
important for many biological applications. The development of automated methods
for nuclear segmentation and classification enables the quantitative analysis of
tens of thousands of nuclei within a whole-slide microscopy image. In situations
of crowded cells, some existing methods can be prone to segmentation errors, such
as falsely merging bordering cells or suppressing valid cell instances due to the
poor approximation with bounding boxes. For this challenge that contains more
than one modality and is diverse in cell shape and color, we use a unet-based
method to tackle this problem. The modified Unet can generate probability maps
and distance maps in one forward step. We then utilize the probability maps and
distance maps to obtain the final cell instance segmentation results.