Advanced parameters

The Advanced parameters window (Fig. 9) provides fine-grained control over Cellects’ algorithmic behavior, enabling researchers to customize analysis pipelines for specific datasets or experimental conditions. These settings extend the flexibility introduced in earlier stages (Data localisation, Image analysis, and Video tracking) by allowing tuning of segmentation logic (e.g., mesh resolution, intensity thresholds), computational performance (parallel processing, memory allocation). For instance, users can optimize for low-noise environments with [Mesh minimal intensity variation] or facilitate high-throughput computations using [Run analysis in parallel]. This section is particularly critical for troubleshooting edge cases (e.g., ambiguous specimen boundaries) or adapting Cellects to non-standard substrates (e.g., heterogeneous growth media). By documenting and reusing these configurations, researchers ensure both technical precision and reproducibility.

Detailed description

Automatically crop images:

Uses initial image detection to crop all images and improve arena/last image detection.

Note

Unselect this option if analysis fails or crashes during image analysis.

Subtract background:

Takes the first image and subtracts it from subsequent images. This can improve or degrade detection depending on dataset characteristics.

All specimens have the same direction:

Select to optimize arena detection for specimens moving move in the same direction.

Checked → Uses motion pattern analysis for arena localization.
Unchecked → Employs standard centroid based algorithm.

Note

Both options work equally when growth is roughly isotropic.
Only works when there is only one specimen per arena

Sliding window segmentation:

Smooth the result of the segmentation to facilitate detection. If checked, each pixel detected as specimen during one of the two previous frames is added to the current frame.

Morphological opening:

Morphological opening first erodes and then dilates all specimens detected during initial segmentation. If checked, this algorithm is applied to every frame after segmentation.

Note

Efficient for removing small noise from the background

Morphological closing:

Morphological opening first dilates and then erodes all specimens detected during initial segmentation. If checked, this algorithm is applied to every frame after segmentation.

Note

Efficient for removing small holes in the detected specimen(s)

Correct errors around initial specimen's position:

Applies an algorithm to correct detection errors near the initial specimen position due to color variations (e.g., from nutrient patches). Technical workflow:

Identifies potential gaps around initial position
Monitors local growth velocity
Fills gaps using growth patterns from adjacent pixels

Note

⚠️ Not recommended if the substrate has the same transparency everywhere (i.e. no differencebetween starting and growth regions).

Prevent fast growth near periphery:

During video analysis, prevents false specimen detection at arena borders by filtering rapid periphery growth.

Checked → Exclude fast -moving detections near boundaries
Unchecked → Use standard detection criteria

Connect distant shapes:

Algorithm for connecting disjoint specimen regions in cases where there should be only one connected specimen per arena. This is useful when the specimen's heterogeneity create wrong disconnections and the detection is smaller than the true specimen. Technical implementation:

Identifies disconnected subregions
Analyzes local growth dynamics
Recreates connections using spatially consistent growth patterns

Note

Increases analysis time substantially.
Only works when there is only one specimen per arena

Appearance size threshold (automatic if checked):

Minimum pixel count threshold for identifying specimen emergence (e.g., bacterial colony formation).

Checked → Automatic threshold calculation.
Unchecked → Manual user -defined threshold.

Appearance detection method:

Selection criteria for initial specimen detection:

Largest: Based on component size metric.
Most central: Based on arena center proximity.

Note

Applicable only to progressively emerging specimens.

Mesh side length:

Pixel dimension for analysis window size.

Note

Must not exceed minimum image dimension

Mesh step:

The size of the step (in pixels) between consecutive rolling window positions.

Note

Must not exceed the mesh side length to ensure full coverage of the image.

Mesh minimal intensity variation:

The minimal variation in intensity to consider that a given window does contain the specimen(s).

Note

This threshold is an intensity value ranging from 0 to 255 (generally small).
Correspond to the level of noise in the background.

Spatio-temporal scaling:

Defines the spatiotemporal scale of the dataset:

Time between images or frames (minutes).
An option to convert areas/distances from pixels to mm/mm².

Keep Cell and Back drawings for all folders:

During initial image analysis, if the user drew cell/back regions to assist detection, this option saves and uses these annotations across all folders. In summary:

Checked → retain annotations for all folders
Unchecked → apply only to current folder

Run analysis in parallel:

Allow the use of more than one core of the computer processor.

Checked → Uses multiple CPU cores to analyze arenas in parallel (faster).
Unchecked → Single core analysis.

Proc max core number:

Maximum number of logical CPU cores to use during analysis. The default value is set to the total number of available CPU cores minus one.

Minimal RAM let free (Go):

Amount of RAM that should be left available for other programs. Setting to 0 gives Cellects all memory, but increases crash risk if other apps are open.

Lose accuracy to save RAM:

For low memory systems:

Converts video from np.float64 to uint8
Saves RAM at the cost of a slight precision loss

Video fps:

Frames per second of validation videos.

Keep unaltered videos:

Keeps unaltered videos (.h5 format) in hard drive.

Checked → Rerunning the same analysis will be faster.
Unchecked → These videos will be written and removed each run of the same analysis.

Note

Large files: it is recommended to remove them once analysis is entirely finalized.

Save processed videos:

Saves lightweight processed validation videos (recommended over unaltered videos). These videos assess analysis accuracy and can be read in standard video players.

Color space combination for video analysis:

Advanced option: Changes the way RGB processing directly in video tracking. Useful for testing new color spaces without (re)running image analysis.

Expected oscillation period:

The period (in minutes) of biological oscillations to detect within the specimen(s). Computation is based on luminosity variations.

Minimal oscillating cluster size:

When looking for oscillatory patterns, Cellects detects connected components that are thickening or slimming synchronously in the specimen(s). This parameter thresholds the minimal size of these groups of connected pixels. This threshold is useful to filter out small noisy oscillations.

Night mode:

Switches the application background between light and dark themes.

Reset all settings:

Useful when the software freezes with no apparent reason. To reset all settings, it removes the config file in the current folder as well as the config file in the software folder. Then, it retrieves and saves the default parameters.