Zeros output for coordinateTransforms time scale

[krokicki]

I'm trying to convert OME-TIFF images which are 2d whole slide image exported by the TissueGnostics microscope software. I wanted to use bioformats2raw to convert these to OME-Zarr for viewing in Neuroglancer. The conversion works but produces .zattrs metadata like this:

{
  "multiscales" : [ {
    "metadata" : {
      "method" : "loci.common.image.SimpleImageScaler",
      "version" : "Bio-Formats 7.3.0"
    },
    "axes" : [ {
      "unit" : "second",
      "name" : "t",
      "type" : "time"
    }, {
      "name" : "c",
      "type" : "channel"
    }, {
      "unit" : "micrometer",
      "name" : "z",
      "type" : "space"
    }, {
      "unit" : "micrometer",
      "name" : "y",
      "type" : "space"
    }, {
      "unit" : "micrometer",
      "name" : "x",
      "type" : "space"
    } ],
    "name" : "Slide 1 of 4_Region 002_channels.tiff",
    "datasets" : [ {
      "path" : "0",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 0.335762932896614, 0.335762932896614 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "1",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 0.671525865793228, 0.671525865793228 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "2",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 1.343051731586456, 1.343051731586456 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "3",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 2.686103463172912, 2.686103463172912 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "4",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 5.372206926345824, 5.372206926345824 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "5",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 10.744413852691649, 10.744413852691649 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "6",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 21.488827705383297, 21.488827705383297 ],
        "type" : "scale"
      } ]
    }, {
      "path" : "7",
      "coordinateTransformations" : [ {
        "scale" : [ 0.0, 1.0, 1.0, 42.977655410766594, 42.977655410766594 ],
        "type" : "scale"
      } ]
    } ],
    "version" : "0.4"
  } ],
  "omero" : {
    "channels" : [ {
      "color" : "0000FF",
      "coefficient" : 1,
      "active" : true,
      "label" : "DAPI",
      "window" : {
        "min" : 0.0,
        "max" : 65535.0,
        "start" : 0.0,
        "end" : 65535.0
      },
      "family" : "linear",
      "inverted" : false
    }, {
      "color" : "FF8000",
      "coefficient" : 1,
      "active" : true,
      "label" : "CY3",
      "window" : {
        "min" : 0.0,
        "max" : 65535.0,
        "start" : 0.0,
        "end" : 65535.0
      },
      "family" : "linear",
      "inverted" : false
    }, {
      "color" : "FF00FF",
      "coefficient" : 1,
      "active" : true,
      "label" : "CY5",
      "window" : {
        "min" : 0.0,
        "max" : 65535.0,
        "start" : 0.0,
        "end" : 65535.0
      },
      "family" : "linear",
      "inverted" : false
    } ],
    "rdefs" : {
      "defaultT" : 0,
      "model" : "color",
      "defaultZ" : 0
    }
  }
}

This fails to load in Neuroglancer with the following error:
"Expected positive finite floating-point number, but received: 0".

This is because the coordinate transforms collapse the time axis to 0 for every scale level. If I change the time axes from 0.0's to 1.0's it loads without issue.

Is there any way to get around this issue? Is the bug in bioformats2raw, Neuroglancer, or TissueGnostics?

[sbesson]

@krokicki as per https://ngff.openmicroscopy.org/0.4/#multiscale-md and notably

If scaling information is not available or applicable for one of the axes, the value MUST express the scaling factor between the current resolution and the first resolution for the given axis, defaulting to 1.0 if there is no downsampling along the axis

I would not expect a 0.0 value in the scale array of the generated OME-Zarr dataset. Could you share a representative OME-TIFF allowing to reproduce this issue?

[krokicki]

Hi @sbesson, here is one of the files which has the issue:

https://s3.janelia.org/scicompsoft/T3/Slide%204%20of%203_Region%20001_channels.tiff

Thanks!

[sbesson]

Thanks for the sample file. Looking at the metadata, the issue comes from

    <Pixels BigEndian="false" DimensionOrder="XYCZT" ID="Pixels:0" Interleaved="false" PhysicalSizeX="0.335762932896614" PhysicalSizeXUnit="µm" PhysicalSizeY="0.335762932896614" PhysicalSizeYUnit="µm" PhysicalSizeZ="1" PhysicalSizeZUnit="µm" SignificantBits="16" SizeC="3" SizeT="1" SizeX="3720" SizeY="5634" SizeZ="1" TimeIncrement="0" TimeIncrementUnit="s" Type="uint16">

and more specifically the 0 value for the Pixels TimeIncrement which is translates into the scale value for the time axis in the OME-Zarr.

This is an scenario which arises from the combination of several metadata values

• SizeT=1. In OME-NGFF, this could be represented as a 5D array (tczxy) with singleton dimensions, which is the current assumption of bioformats2raw, or a 3D array (cyx). In the latter case, there would be no additional issue as the scale array would include no time element
• PixelsTimeIncrement=0. From the perspective of the OME model, this is valid as https://www.openmicroscopy.org/Schemas/Documentation/Generated/OME-2016-06/ome_xsd.html#Pixels_TimeIncrement is a floating-point value. Given that SizeT=1, my expectation would have been that this attribute would have been omitted from the OME metadata as it is not meaningful. In that case, the converter would have defaulted to 1.0 values in the scale array
• "scale" : [ 0.0, 1.0, 1.0, 0.671525865793228, 0.671525865793228 ] while my initial feeling is this should not expected, I am now reconsidering this in the light of "If scaling information is not available or applicable for one of the axes". In this scenario, the time scaling information is available in the OME-XML metadata and there is no statement that unambiguously disallows a 0.0 value.

Overall, I think this use case could be handled more gracefully in the writer (bioformats2raw), the reader (Neuroglancer) and possibly clarified in the specification. I have opened #281 to avoid writing 0.0 values as a time scale value. See also #279 for ongoing efforts towards collapsing singleton dimensions during conversion