Signals and events

Prerequisites

Perform segmentation, tracking, and measurements for either target or effector cells. Select a single position.

Overview

After measuring single-cell features over time, the next step is to characterize dynamic behaviors — detecting when and whether specific events occur for each cell. Celldetective offers two complementary strategies for this: deep learning signal analysis and threshold-based event detection.

Deep-learning signal analysis

Celldetective provides a zoo of deep-learning models that take single-cell signal traces as input and predict an event class and time of event for each cell. These models work similarly to segmentation models — select one, map your measurement columns to the model’s expected inputs, and submit.

For a detailed list of signal mapping parameters, see the Signal Analysis Reference.

Threshold-based event detection

As an alternative to deep learning, you can define feature-based classification rules (e.g., PI_intensity_mean > 500) that produce a binary signal per cell. For tracked cells with time-correlated events, a sigmoid is fitted to extract the event time. The quality of the fit is assessed by an \(R^2\) score.

Single-cell signal viewer

Celldetective ships a powerful viewer for exploring single-cell signals and manually annotating events. This tool allows you to:

Visualize single-cell signal traces (intensity, morphology) synchronized with the movie.
Manually annotate event times and classes.
Curate datasets to train deep-learning event detection models.

signal_annotator — Application on an ADCC system of MCF-7 breast cancer cells co-cultured with human primary NK cells.

The viewer displays the movie with cell centroids marked by their current event status. Clicking a cell reveals its full temporal signal trace.