Segmentation Task Definition

The segmentation_tasks object in the segmentation algorithm file is a list of tasks. Each task has the following structure:

{
    "task_id": 0,
    "segmentation_family": "Cellpose",
    "entity_types_detected": [ ... ],
    "z_layers": [ ... ],
    "segmentation_properties": { ... },
    "task_input_data": [ ... ] ,
    "segmentation_parameters": { ... },
    "polygon_parameters": { ... },
}

task_id

Each task has an integer index, starting from zero called the task_id. The task_id is used in the generation of the EntityID to ensure uniqueness.

Note

The Vizgen Post-processing Tool vpt only supports 9 tasks in a single segmentation algorithm. If more tasks are needed, users are encouraged to run vpt multiple times and combine the segmentation outputs post hoc.

segmentation_family

Each task has a segmentation_family. This variable specifies what type of segmentation will be performed and what parameters must be provided. The currently supported families are:

• Watershed

• Cellpose

entity_types_detected

A task may detect one or more entities based on entity_types_detected. This is useful for running multiple tools for detecting an Entity (e.g. Cells) and combining the output. Supported Entity types are "cell" and "nucleus."

z_layers

The z_layers variable is a list of which z-index images should be used in this segmentation task. If the type of segmentation is 2D and multiple z-indexes are entered, segmentation will be performed on each z-layer sequentially and the Entities will be merged into 3D objects at the end of the task. If multiple z_layers are selected with 2D segmentation, a heuristic will be applied to fill in any gaps between segmented layers. 2D segmentation is applied to all z-layers in the experiment by default without needing to explicitly specify the behavior.

Properties and Parameters

Both segmentation_properties and segmentation_parameters are specific to the segmentation_family and are described in their own sections.

task_input_data

The task_input_data list is a list of image channels and per-channel pre-processing steps. This tool allows the application of intensity normalization, down-sampling, and/or blurring before performing cell segmentation.

Normalization Options

Filter name: "normalize"

Types:

• default (default if no parameters are specified)

  default normalization is min-max range normalization. It does not accept any parameters

• CLAHE

  Contrast Limited Adaptive Histogram Equalization (CLAHE) is a normalization method for emphasizing local contrast in an image at the expense of altering global brightness. CLAHE accepts two parameters:

  • clip_limit (range: 0 - 1)

  • filter_size (2 element list, units of pixels)

  For more information, see the Scikit Image Documentation

Usage:

"image_preprocessing": [
    {
        "name": "normalize",
        "parameters": {
            "type": "CLAHE",
            "clip_limit": 0.01,
            "filter_size": [
                100,
                100
            ]
        }
    }
]

Blur Filter Options

Filter name: "blur"

Types:

• average (default if no parameters are specified)

• median

• gaussian

All blur filter types accept a kernel size parameter in units of pixels and are implemented using OpenCV. If the size is not specified it takes the default value 5px.

Usage:

"image_preprocessing": [
    {
        "name": "blur",
        "parameters": {
            "type": "gaussian",
            "size": 21
        }
    }
]

Downsample Options

Filter name: "downsample"

The downsample filter reduces the size of the images during segmentation to decrease processing time. The only parameter passed to the downsample filter is the scale factor, scale (default value: 2).

Usage:

"image_preprocessing": [
    {
        "name": "downsample",
        "parameters": {
            "scale": 2
        }
    }
]

polygon_parameters

The polygon_parameters are used when converting a foreground / background mask into geometric shapes describing the outlines of cells. In order to improve computational performance and avoid artifacts, smoothing and simplification of Entity geometries is recommended, as is filtering by size. All parameters are measured in units of pixels.

Parameters:

• simplification_tol

  The acceptable loss of precision when simplifying cell boundaries. Even a small amount of simplification (2 px) dramatically improves processing time.

• smoothing_radius

  The size of a smoothing operation comparable to morphologically closing and then opening the cell mask using the same structuring element.

• minimum_final_area

  Minimum area of a polygon to retain the cell. Used to filter spurious detections.

Usage:

"polygon_parameters": {
    "simplification_tol": 2,
    "smoothing_radius": 10,
    "minimum_final_area": 500
}