[
International Worm Meeting,
2019]
We have advanced a collaborative machine learning system to solve the challenging problem of identifying cells in C elegans from still or video volumetric microscopy. The distinguishing features of this approach versus prior approaches include: (1) use of a statistical model of cell features that is iteratively improved, (2) generation of probabilistic guesses at cell ID rather than single best-guesses for each cell, (3) tracking of joint probabilities of features within and across cells, and (4) ability to exploit multi-modal features, such as cell position, morphology, reporter intensities, and activity. We have developed a generative spring-mass model to simulate sequences of cell imaging datasets with variable cell positions and fluorescence intensities. We explore various probability models of increasing sophistication, and we have developed a novel, effective algorithm for Bayesian label matching, part of a class of generally intractable combinatorial optimization problems. We find that atlases that track inter-cell positional correlations give higher labeling accuracies than those that treat cell positions independently. Tracking an additional feature type, fluorescence intensity, boosts accuracy relative to a position-only atlas, demonstrating that multiple cell features can be leveraged to improve automated label predictions. We demonstrate use of this open-source system on whole body still and video volumetric imaging in adult hermaphrodite C. elegans, for both single fluorophore and multi-fluorophore expressing worm strains.