How To Make Your Cell Tracker Say "I dunno!"

TL;DR

The paper addresses uncertainty in automated cell tracking under high-throughput imaging by introducing two complementary viewpoints: a Bayesian framing that yields a probabilistic cell-tracking posterior and an accompanying MAP solution, and a classification framing that produces edge-level confidences via per-edge probabilities. It offers a plug-in framework that can wrap any linear assignment-based TbD tracker, including distance-, overlap-, activity-based methods and Transformer-based Trackastra, with techniques such as top-$K$ assignment sampling, feature perturbation, and temperature scaling to calibrate uncertainties. Key contributions include formalizing the cell-tracking posterior, proposing efficient ensemble and perturbation strategies to approximate predictive posteriors, and developing a daughter-based mother classification approach with parental softmax and temperature scaling for calibrated edge confidences and entropy-based uncertainty. Empirical findings show that many trackers are overconfident, particularly at lower frame rates, but that calibration via TS and uncertainty-guided sparsification can improve practical usefulness, enabling reliable automated tracking and informed human-in-the-loop corrections.

Abstract

Cell tracking is a key computational task in live-cell microscopy, but fully automated analysis of high-throughput imaging requires reliable and, thus, uncertainty-aware data analysis tools, as the amount of data recorded within a single experiment exceeds what humans are able to overlook. We here propose and benchmark various methods to reason about and quantify uncertainty in linear assignment-based cell tracking algorithms. Our methods take inspiration from statistics and machine learning, leveraging two perspectives on the cell tracking problem explored throughout this work: Considering it as a Bayesian inference problem and as a classification problem. Our methods admit a framework-like character in that they equip any frame-to-frame tracking method with uncertainty quantification. We demonstrate this by applying it to various existing tracking algorithms including the recently presented Transformer-based trackers. We demonstrate empirically that our methods yield useful and well-calibrated tracking uncertainties.

Figures (2)

• Figure 1: Example of ambiguity in two different assignment solutions $A_1$ and $A_2$ caused by the similar appearance of cells and missing frames. Assignments are color-coded, i.e. cells in frame $t_2$ with the same color as those in $t_0$ are considered daughters of the latter. Standard tracking methods yield point estimates, i.e. they will choose either solution, but not report any uncertainty caused by the ambiguity in choosing the correct mother for the cell marked by the red arrow.
• Figure :