Tracking by weakly-supervised learning and graph optimization for whole-embryo C. elegans lineages

Peter Hirsch, Caroline Malin-Mayor, Stephan Preibisch, Dagmar Kainmueller, Jan Funke

Keywords: Object Tracking, Detection, Microscopy, Graph Optimization, ILP, Deep Learning

TL;DR: We combine deep learning and ILP-based graph optimization to track all cells in whole embryos

Abstract:

The tracking of all nuclei of an embryo in noisy and dense fluorescence microscopy data is a challenging task. We build upon a recent method that combines deep learning to extract candidate solutions with an integer linear program (ILP) to select the most likely tracks. We present extensions of this method to specifically address the following challenging properties of C. elegans embryo recordings: (1) Relatively many cell divisions compared to benchmark recordings of other organisms, and (2) the presence of polar bodies, which look similar to cell nuclei and are thus easily mistaken as such. To cope with (1), we devise and incorporate a learnt cell division detector. To cope with (2), we devise and incorporate a learnt polar body detector. We further extend the method to allow for automated ILP hyperparameter tuning via a structured SVM, thus alleviating the need for tedious manual set-up of a respective grid search.

At the time of submission, our method heads the leaderboard of the cell tracking challenge on the Fluo-N3DH-CE C. elegans embryo dataset. Furthermore, we report an extensive quantitative evaluation of our method on two additional C. elegans datasets, namely a set of 3 fully annotated confocal embryo recordings, and a set of 3 fully annotated lightsheet embryo recordings. We will make these datasets public to serve as an extended benchmark for future method development. To gauge the practical impact of our method, we include the software Starrynite as baseline. Starrynite is commonly employed by biologists for the study of C. elegans. Our results suggest considerable improvements, especially in terms of the correctness of division event detection and the number of fully correct tracks.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: methodological development

Primary Subject Area: Segmentation

Secondary Subject Area: Learning with Noisy Labels and Limited Data

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2208.11467/code)