OrganoidTracker: Efficient cell tracking using machine learning and manual error correction

Abstract

Time-lapse microscopy is routinely used to follow cells within organoids, allowing direct study of division and differentiation patterns. There is an increasing interest in cell tracking in organoids, which makes it possible to study their growth and homeostasis at the single-cell level. As tracking these cells by hand is prohibitively time consuming, automation using a computer program is required. Unfortunately, organoids have a high cell density and fast cell movement, which makes automated cell tracking difficult. In this work, a semi-automated cell tracker has been developed. To detect the nuclei, we use a machine learning approach based on a convolutional neural network. To form cell trajectories, we link detections at different time points together using a min-cost flow solver. The tracker raises warnings for situations with likely errors. Rapid changes in nucleus volume and position are reported for manual review, as well as cases where nuclei divide, appear and disappear. When the warning system is adjusted such that virtually error-free lineage trees can be obtained, still less than 2% of all detected nuclei positions are marked for manual analysis. This provides an enormous speed boost over manual cell tracking, while still providing tracking data of the same quality as manual tracking.

Multiscale light-sheet organoid imaging framework

“…Although with high degree of accessibility, performing live imaging of organoid growth remains a challenge, as it typically requires not only microscopy techniques capable of stable long-term imaging of several samples simultaneously, but also dedicated analysis and processing pipelines that can cope with complex imaging data. Nonetheless, previous work on the live recording of organoid dynamics has focused either on specific biological questions 6,8,9 or on specific isolated tools 10 without a more generalised yet in-depth approach on light-sheet imaging and data analysis. The only current work which aimed at creating a light-sheet organoid imaging platform 11 focused mainly on the determination of culture-wide heterogeneities through a combination of both light-sheet and wide-field techniques.…”

Cell Tracking for Organoids: Lessons From Developmental Biology

“…Key to scaling up such a hybrid approach from a limited number of lineages to entire organoids is to incorporate algorithms that automatically identify possible errors and allow for efficient manual correction of these errors. Recently, we developed such a hybrid approach to perform lineage tracking for whole intestinal organoids ( Kok et al, 2020 ).…”

A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction

“…The resulting segmentation errors can be classified as False Positives (FP), False Negatives (FN), over-segmentation, under-segmentation, and wrong partitioning of touching cells [43].To handle such segmentation errors in tracking by detection methods, two strategies exist: 1) Generating overlapping segmentation masks and selecting the final set of segmentation masks in the tracking step [31,32,34,36,42,44]. 2) Using non overlapping segmentation masks and detecting and correcting segmentation errors [24,28,30,33,35,39,40,[45][46][47]. The first strategy is computationally expensive as several segmentation hypothesis are competing.…”

“…The first strategy is computationally expensive as several segmentation hypothesis are competing. For the second strategy, semi-automated methods with manual data curation [45,46,48] and automated methods [24,28,30,33,35,39,40,47] have been proposed.While semi-automated methods need manual effort for error correction, automated segmentation correction approaches often require a learning step. For instance, classifiers that estimate the number of objects per segmentation mask are trained [30,33], where no objects correspond to FPs, and more than one object to an under-segmentation error.…”

“…The first strategy is computationally expensive as several segmentation hypothesis are competing. For the second strategy, semi-automated methods with manual data curation [45,46,48] and automated methods [24,28,30,33,35,39,40,47] have been proposed.…”