About:

This module is for segmenting 3D membrane tagged cell images. The workflow is as follows: 3D UNet to output the probability map of cell boundary, 3D watershed to get the initial cell label, and 3D CRF to refine the segmentation label.

Hyperparameters

• Minimum distance: In the seeds detection for watershed, the minimum possible distance between two seeds. The lower the value is, the more likely to have oversegmentation.
• Label Threshold: The minimum possible volumn of cells to avoid segmenting intercellular space as an individual cell.
• Threshold: The bilateral standard deviation (threshold) for CRF define how much intensity-homogeneity is required within a region. Higher values allow greater variations under the same label.

Paper: Accurate 3D Cell Segmentation Using Deep Features and CRF Refinement.

Input

The input should be 3D volumentric cell membrane tagged image stack in TIFF format (z,x,y).

Outputs

• Cell Number: The number of cells in each image stack.
• Cell Volume: The size of each cell in terms of number of voxels.
• Cell Center: The centroid of each cell.
• Cell Coordinates: The surface coordinate of each cell.
• Adjacent Table: The graph represents the neighber of each cell.
• 3-way Conjunction Points: The touch points of 3 neighboring cells.