Automated Reconstruction of Whole-Embryo Cell Lineages by Learning from Sparse Annotations

Abstract

We present a method for automated nucleus identification and tracking in time-lapse microscopy recordings of entire developing embryos. Our method combines deep learning and global optimization to enable complete lineage reconstruction from sparse point annotations, and uses parallelization to process multi-terabyte light-sheet recordings, which we demonstrate on three common model organisms: mouse, zebrafish, Drosophila. On the most difficult dataset (mouse), our method correctly reconstructs 75.8% of cell lineages spanning 1 hour, compared to 31.8% for the previous state of the art, thus enabling biologists to determine where and when cell fate decisions are made in developing embryos, tissues, and organs.

Tracking by Weakly-Supervised Learning and Graph Optimization for Whole-Embryo C. elegans lineages

“…Our method extends the tracking-by-detection approach Linajea [14]. We briefly review Linajea, and then describe our extensions in detail.…”

“…1: Method overview: We use a 4d U-Net to predict cell candidates and movement vectors. These are used to construct a candidate graph G with node and edge scores g s as in [14]. We propose to integrate learnt cell state scores cs s .…”

“…Graph G, feature matrix S, weights w and a set of feasibility constraints form an ILP that yields the cell lineage. We propose to find w via a structured SVM (sSVM); proposed changes highlighted in magenta; figure adapted from [14].…”

“…Linkage can also be achieved in a globally optimal manner by means of combinatorial optimization, as in the competitive method Baxter [13] (former rank 2 on the CTC C. elegans benchmark), which employs the Viterbi algorithm in the linkage step. Related methods can also directly yield an optimal feasible cell lineage tree, in the face of over-and underdetections [22], as well as for an overcomplete set of candidate detections [9,14]. To modernize the detection step, Cao et al [2] replaced the classical detection step of Starrynite by neural network-based cell segmentation.…”

“…We propose a method that unifies the individual advantages of Baxter [13] and Elephant [23]. To this end we build upon Linajea [14], a recent method that combines deep learning and combinatorial optimization for cell tracking. We propose extensions of Linajea to capture properties specific to recordings of the model organism C. elegans, namely relatively many cell divisions, and the presence of polar bodies which look similar to nuclei.…”