Cell Detection with Star-convex Polygons

TL;DR

This paper introduces a novel deep learning method for cell detection that uses star-convex polygons to accurately localize cell nuclei in microscopy images, especially in crowded scenarios.

Contribution

It proposes a new shape representation for cell detection using star-convex polygons, eliminating the need for shape refinement and improving accuracy in crowded cell images.

Findings

Effective on synthetic and real microscopy datasets

Outperforms bounding box-based methods in crowded scenarios

Accurate cell instance localization without shape refinement

Abstract

Automatic detection and segmentation of cells and nuclei in microscopy images is important for many biological applications. Recent successful learning-based approaches include per-pixel cell segmentation with subsequent pixel grouping, or localization of bounding boxes with subsequent shape refinement. In situations of crowded cells, these 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. To overcome these issues, we propose to localize cell nuclei via star-convex polygons, which are a much better shape representation as compared to bounding boxes and thus do not need shape refinement. To that end, we train a convolutional neural network that predicts for every pixel a polygon for the cell instance at that position. We demonstrate the merits of our approach on two synthetic datasets and one challenging dataset of diverse fluorescence microscopy images.