The following file formats are compatible for use with the import-segmentation pipeline:

• Labeled mask in TIFF (32-bit) or NumPy NPY format
• Polygons in GeoJSON format (FeatureCollection type)
• Baysor output CSV and GeoJSON file formats
• Xenium Onboard Analysis cells.zarr.zip file: the nucleus masks are used as input if --nuclei=cells.zarr.zip and can be expanded with --expansion-distance. The cell masks are used as input if --cells=cells.zarr.zip.

The polygon GeoJSON should be exported as a FeatureCollection. The cells and nuclei can be saved in the same (as shown below in QuPath's output format) or separate GeoJSON files. The format should look similar to this example:

Copy
{
   "type": "FeatureCollection",
   "features": [
     {
       "type": "Feature",
       "id": "12059192-3b27-4438-96dc-97b41ca84717",
       "geometry": {
         "type": "Polygon",
         "coordinates": [
           [
             [3418.94, 2],
             [3414.62, 3.85],
             [3414.12, 22.65],
             [3415.06, 27.26],
             [3420.1, 35.15],
             [3428.03, 40.22],
             [3437.01, 42.94],
             [3440.69, 45.87],
             [3445.32, 46.71],
             [3468.85, 46.71],
             [3477.88, 44.35],
             [3534.35, 44.31],
             [3546.79, 37.83],
             [3552.28, 30.11],
             [3552.94, 25.45],
             [3552.93, 6.63],
             [3550.68, 2.49],
             [3546, 2],
             [3418.94, 2]
           ]
         ]
       }
     }
   ]
}

Segmentation results from QuPath are compatible with Xenium Ranger and have these specifications:

• The feature objectType should be cell, as features with a non-cell objectType will be ignored (i.e., annotations).
• The --nuclei argument will use the nucleusGeometry polygon if it exists in the GeoJSON, otherwise it will use the geometry polygon. The --cells argument will use the geometry polygon.
• QuPath exports nucleus and cell segmentation results in one file, so the same GeoJSON file should be specified for both --cells and --nuclei. The format should look similar to this example:

Copy
{
  "type":"FeatureCollection",
  "features": [
    {
      "type":"Feature",
      "id":"fd0c3d4e-6146-427d-9696-97fbe7adb63d",
      "geometry":{
        "type":"Polygon",
        "coordinates":[
          [
           [4348.52, 0],
           [4344.02, 1.37],
           [4341.18, 10.23],
           [4341.8, 24.31],
           [4346.73, 32.28],
           [4353.27, 38.99],
           [4361.18, 44],
           [4379.94, 44.48],
           [4388.64, 40.97],
           [4396.11, 35.26],
           [4404.71, 21.36],
           [4404.27, 2.55],
           [4400.29, 0.06],
           [4348.52, 0]
           ]
          ]
      },
      "nucleusGeometry":{
        "type":"Polygon",
        "coordinates":[
          [
           [4373.91, 4.81],
           [4366.61, 9.07],
           [4364.28, 12.61],
           [4364.3, 16.85],
           [4370.37, 22.75],
           [4378.35, 20.24],
           [4382.75, 13.06],
           [4380.87, 9.26],
           [4373.91, 4.81]
           ]
          ]
      },
      "properties":{
        "objectType":"cell"
      }
    }
  ]
}

Segmentation results should match output formats from Baysor v0.6 to be compatible with Xenium Ranger and have the specifications listed below.

Baysor segmentation CSV:

• The CSV must have the following columns: transcript_id, cell, is_noise (order does not matter). The other default columns from Baysor are optional.

Here is an example of the required columns in CSV format:

Copy
transcript_id,cell,is_noise
281474976710656,CRc17aaabcd-3,false
281474976710657,CRc17aaabcd-3,false
281474976710658,CRc17aaabcd-3,false
[ ... ]

Baysor segmentation GeoJSON:

• Should be of type GeometryCollection, where each geometry is of type Polygon.
• The Polygon should contain an extra field called cell to identify which cell the polygon corresponds to.
• Each cell in the GeoJSON must have at least one transcript assigned to it in the transcript assignment CSV. Filtering out Q-Score < 20 transcripts and negative controls prior to running any transcript-based segmentation method is recommended. However, unlike Xenium Ranger v2.0.1, Xenium Ranger v3.0 can handle cells with only low-quality transcripts without mixing up a cell's transcripts. See this Knowledge Base article for more information.

Here is an example of the GeoJSON format:

Copy
{
  "geometries": [
    {
      "coordinates": [
        [
          [15.891722, 21.441122],
          [13.95383, 21.565407],
          [14.233208, 25.427168],
          [15.927658, 26.113512],
          [17.464699, 25.211208],
          [16.642977, 22.83285],
          [15.891722, 21.441122]
        ]
      ],
      "type": "Polygon",
      "cell": 5
    }
  ],
  "type": "GeometryCollection"
}