CellDIVE pipeline

Prepares data and does 2D segmentation of CellDIVE using deepcell method. More information about DeepCell and the associated Mesmer algorithm can be found in the associated publication here.

This method was selected based on its performance in a number of metrics calculated without the use of a human segmented reference image in the process described in this publication

Usage example

cwltool pipeline.cwl subm.yaml

Requires meta.yaml with names of channels that will be used for segmentation of cell and nucleus compartments.

The expected input directory structure:

HuBMAP_OME/
├── region_001
│     ├── S20030077_region_001.ome.tif
│     ├── S20030078_region_001.ome.tif
│     │              ...
│     └── S20030105_region_001.ome.tif
└── region_00N
      ├── S20030077_region_00N.ome.tif
      ├── S20030078_region_00N.ome.tif
      │             ...
      └── S20030105_region_00N.ome.tif

The output structure:

pipeline_output/
├── expr
│   ├── reg001_S20030077_region_001_expr.ome.tiff
│   ├── reg002_S20030078_region_002_expr.ome.tiff
│   │                  ...
│   └── reg00N_S20030105_region_00N_expr.ome.tiff       
└── mask                                                                                  
    ├── reg001_S20030077_region_001_mask.ome.tiff
    ├── reg002_S20030078_region_002_mask.ome.tiff               
    │                  ...                   
    └── reg00N_S20030105_region_00N_mask.ome.tiff