Settings & Parameters

This reference page lists the configuration parameters for various Celldetective modules.

Segmentation Data Import

These parameters appear in the Upload Model window when importing a pretrained model.

General Settings (All Models)

• Input spatial calibration: The pixel resolution (in microns) of the images the model was trained on.

• Channel Mapping: Map the model’s expected inputs (e.g., “Channel 1”, “Cyto”, “Nuclei”) to your experiment’s channels. Select -- to ignore.

• Normalization:

  • Mode: Check for percentile-based standard scaling (0-1). Uncheck for raw values.

  • Clip: Check to clip values outside the chosen percentile range.

  • Range: Min/max percentiles for normalization (e.g., 1.0 - 99.8).

Cellpose Specifics

• Cell Diameter [px]: The average object diameter in the training data. If set to 30.0 (default), Cellpose assumes standard scaling.

• Cellprob Threshold: Threshold for the confidence map (default 0.0). Lower values increase sensitivity.

• Flow Threshold: Threshold for flow error (default 0.4). Lower values enforce stricter shapes.

Segmentation Runtime Settings

These parameters appear when applying a generalist model.

StarDist (Generalist)

• Channel Selection: Map specific experiment channels (e.g., Nuclei) to the model’s input.

Cellpose (Generalist)

• Channel Mapping: Select “Cytoplasm” and “Nuclei” channels.

• Diameter [px]: Expected cell diameter. Use the button to open the Interactive Diameter Estimator.

• Flow/Cellprob Thresholds: Adjust detection sensitivity and shape constraints on the fly.

Tracking Settings

Accessible via the button in the Tracking module.

Trackers

• bTrack: Bayesian tracker using Kalman filters and visual features.

• trackpy: Particle tracker based on Crocker-Grier.

  • Search range [px]: Max movement distance per frame.

  • Memory [frames]: Max frames a particle can disappear.

Feature Extraction

• Morphological features & Intensity:

  • Standard: area, eccentricity, solidity, perimeter, intensity_mean, intensity_max, intensity_min, etc.

  • Advanced: major_axis_length, minor_axis_length, orientation, extent, euler_number, feret_diameter_max.

  • Custom: Any allowed function from skimage.measure.regionprops.

• Haralick Texture Features:

  • Target channel: Channel to analyze (must be one of the loaded channels).

  • Distance: Pixel distance for GLCM calculation (default 1).

  • # gray levels: Number of intensity bins for quantization (default 256).

  • Scale: Downscaling factor (0-1) to speed up computation.

  • Normalization:

    • Percentile Mode: Normalize intensities between min/max percentiles (e.g., 1% - 99.9%).

    • Absolute Mode: Normalize intensities between fixed pixel values.

Post-Processing

Setting	Description
Min. tracklength	Filter out tracks shorter than this number of frames.
Remove tracks… (Start)	Remove tracks that do not start at the first frame.
Remove tracks… (End)	Remove tracks that do not end at the last frame.
Interpolate gaps	Fill missing detections (gaps) within a track using linear interpolation.
Extrapolate (Pre)	Sustain the first detection’s position backwards to the start of the movie.
Extrapolate (Post)	Sustain the last detection’s position forwards to the end of the movie.

Neighborhood Measurement Settings

Accessible when selecting Neighborhood in Measurements.

Population Configuration

• Reference / Neighbor: Select the two populations to analyze (can be the same for self-neighborhood).

• Filters:

  • Status: Restrict analysis to cells with a specific status (e.g., “Alive”, “Positive”).

  • Not: Check the “Not” button () to invert the status selection (e.g., Select “Alive” and check “Not” to target “Dead” cells).

  • Event Time: Correlate measurements with a specific event (e.g., t_death). This creates event-aligned neighborhood metrics.

• Cumulated Presence: If checked, computes the total duration (in frames or time) that a neighbor has been present within the defined threshold.

Measurement Types

• Distance Threshold: Detects neighbors within a fixed radial distance from the cell centroid.

  • Distance [px]: The radius of the neighborhood circle. Can add multiple distances.

• Mask Contact: Detects neighbors whose boundaries are within a specific proximity.

  • Distance [px]: The maximum distance between cell boundaries to be considered “in contact” (often 0 for touching or small positive value for near-contact).

General Options

• Clear Previous: If checked, removes all previously computed neighborhood columns from the data tables before saving new ones. Essential when re-running analysis with different parameters to avoid clutter.

Survival Analysis Settings

Accessible via Analyze > Plot Survival.

Data Selection

• Population: Target cell population.

• Time of Reference: Start point (\(T=0\), e.g., t_firstdetection).

• Time of Interest: End event (e.g., t_death).

Filtering

• Query: Pandas query string helper (e.g., TRACK_ID > 10).

• Cut obs. time [min]: Censoring threshold.

Visualization

• Time calibration: Frames-to-minutes conversion.

• Cmap: Colormap for curves.

Single Cell Measurements

Accessible via the Analyze > Measure tab.

Isotropic Measurements

Measurements taken within circular or ring-shaped ROIs centered on the cell.

• Radii [px]: List of radii (e.g., 10) or rings (e.g., 10-20) defining the ROIs.

• Operations: Statistical operations to perform within the ROI (mean, std, sum, median, min, max).

Contour Measurements

Measurements taken within a band relative to the cell boundary.

• Distances [px]: List of distances from the mask edge. Positive values are inside (erosion), negative values are outside (dilation). Pairs (e.g., (0, 5)) define a band.

Spot Detection

Detection of intracellular spots (e.g., FISH probes) using Laplacian of Gaussian.

• Channel: Target channel for spot detection.

• Diameter [px]: Expected diameter of the spots.

• Threshold: Sensitivity threshold for detection.

• Preprocessing: filters to apply before detection (e.g., smooth, denoise).

Segmentation Model Training

Accessible via Train > Segmentation Model.

Model Selection

• Model Type:

  • StarDist: Best for round/convex objects (nuclei).

  • Cellpose: Best for complex shapes and cytoplasm.

• Pretrained Model: Initialize weights from an existing model (Generic or Custom).

• Model Name: Unique name for the new model.

Training Data

• Training Data: Folder containing images and masks (e.g., from an annotated experiment).

• Include Dataset: Select a built-in dataset to augment training.

• Augmentation Factor: Multiplier for data augmentation (rotation, flip, zoom). Default 2.0.

• Validation Split: Fraction of data reserved for validation (e.g., 0.2).

Hyperparameters

• Learning Rate: Step size for the optimizer (StarDist default: 0.0003, Cellpose default: 0.01).

• Batch Size: Number of images per training step (default 8).

• Epochs: Number of training iterations (StarDist default: 100-500, Cellpose default: 100-10000).

Experiment Configuration (config.ini)

The config.ini file is created automatically when you set up a new experiment (see How to create a new experiment). It uses the standard INI format and is located at the root of the experiment folder. Below is a complete reference of every section and key.

[Populations]

Declares which cell populations are included in the experiment.

Key	Type	Description
populations	string	Comma-separated list of population names (e.g. targets,effectors). These names match the population folders created inside each position directory.

[MovieSettings]

Image-acquisition and stack geometry parameters.

Key	Type	Description
pxtoum	float	Spatial calibration: how many micrometres one pixel represents (default 1.0).
frametomin	float	Temporal calibration: the interval in minutes between two consecutive frames (default 1.0). For single-time-point data, leave at 1.0.
len_movie	int	Number of frames in the movie. Used as a fallback when automatic frame-count extraction fails. For variable-length stacks, set a conservative (lower) estimate.
movie_prefix	string	Filename prefix that stack files must start with to be loaded (e.g. Experiment). Leave blank if filenames have no common prefix.
shape_x	int	Image width in pixels (default 2048).
shape_y	int	Image height in pixels (default 2048).

[Channels]

Maps channel names to their stack index (0-based). Each key is a channel name and each value is the integer index of that channel in the multi-channel stack, or nan if the channel is not present.

Example

[Channels]
brightfield_channel = 0
adhesion_channel = 1
fitc_channel = 2
cy5_channel = nan

Built-in channel names include brightfield_channel, live_nuclei_channel, dead_nuclei_channel, effector_fluo_channel, adhesion_channel, fluo_channel_1, fluo_channel_2. Custom channel names can be added during experiment creation.

[Labels]

Per-well biological condition labels. Each value is a comma-separated list whose length equals the number of wells in the experiment.

Key	Type	Description
cell_types	string	Cell type for each well (e.g. NK,NK,T-cell,T-cell).
antibodies	string	Antibody used in each well (e.g. anti-CD4,anti-CD4,none,none).
concentrations	string	Antibody or drug concentration for each well (e.g. 0,100,0,100).
pharmaceutical_agents	string	Pharmaceutical agent applied in each well (e.g. none,dextran,none,dextran). Fields can be left blank (defaults to well index).

[Metadata]

Additional experiment-level metadata.

Key	Type	Description
concentration_units	string	Unit for concentration values in [Labels] (default pM).

Full example

[Populations]
populations = targets,effectors

[MovieSettings]
pxtoum = 0.325
frametomin = 3.0
len_movie = 120
movie_prefix = Experiment
shape_x = 2048
shape_y = 2048

[Channels]
brightfield_channel = 0
adhesion_channel = 1
fitc_channel = 2

[Labels]
cell_types = NK,NK,T-cell,T-cell
antibodies = anti-CD16,anti-CD16,none,none
concentrations = 0,100,0,100
pharmaceutical_agents = none,dextran,none,dextran

[Metadata]
concentration_units = pM

Preprocessing Protocols

Accessible via the Preprocessing module.

General Correction Settings

• Operation:

  • Subtract: Subtract the estimated background from the image.

  • Divide: Divide the image by the background (flat-field correction).

• Clip: (Subtract mode only) Clip negative values to zero after subtraction.

• Offset: Camera black level/offset. Subtracted prior to background estimation.

• Interpolate NaNs: Fill missing or NaN pixels using neighboring values.

Background Correction

• Model Fit: Fits a 2D surface (plane/paraboloid) to the background.

  • Model type: paraboloid (best for curved illumination) or plane (best for simple gradients).

  • Threshold: Standard deviation threshold to exclude cells/objects from the fit.

  • Downsample: Factor to downsample images for faster surface fitting (default: 10).

• Model Free: Computes a median background image from multiple positions or timeframes.

  • Stack mode:

    • timeseries: Estimates background from a range of frames in the current position.

    • tiles: Estimates background across all positions/tiles (best for global background).

  • Time range: Specific frames to use for estimation (only in timeseries mode).

  • Threshold: Standard deviation threshold to mask cells during estimation.

  • Optimization:

    • Optimize for each frame: If checked, performs a linear regression to adjust the background level per-frame.

    • Coef. range: Range of scaling factors allowed during optimization (e.g., 0.95 - 1.05).

    • Nbr of coefs: Number of values to test within the coefficient range.

Local Correction

• Distance: The radial distance (in pixels) from the cell mask boundary used to estimate local background.

• Model: mean or median of intensity within the boundary band.

Channel Offset

• Shift (h)/(v): Pixel shift (horizontal and vertical) to align the target channel with the reference.

• Viewer: Use the button to open the Offset Viewer. Use arrow keys to visually align the channels.

Signal Analysis

Signal Mapping

Configuration window for Deep Learning signal models.

• Required Inputs (Left): The specific signals expected by the model (e.g., “Nuclei Intensity”).

• Available Columns (Right): The columns from your measurement table to map to these inputs.

Event Annotation

Configuration for the Single Cell Signal Annotator.

• Image Mode:

  • Grayscale: Single channel visualization.

  • Composite: RGB overlay (requires channel selection and per-channel normalization).

• Rescaling: Downscaling fraction (e.g., 0.5) to reduce memory usage during animation.

• Time Interval: Playback speed (milliseconds between frames).