Autominerva could not read metadata from backsub generated ome TIFF image

[yujxiao]

The pipeline runs fine with either backsub and no --viz false, or without backsub and --viz true. When running both backsub and minerva visualization, the error happens as below and the ome tiff generated is incomplete.

ERROR ~ Error executing process > 'viz:autominerva (1)'

Caused by:
Process viz:autominerva (1) terminated with an error exit status (1)

Command executed:

python /app/story.py --in rack-01-well-A01-roi-001-exp-1_backsub.ome.tif --m markers_bs.csv --out story.json
python /app/minerva-author/src/save_exhibit_pyramid.py rack-01-well-A01-roi-001-exp-1_backsub.ome.tif story.json rack-01-well-A01-roi-001-exp-1_backsub

Command exit status:
1

Command output:
Using marker names from markers_bs.csv

Command error:
opening image: rack-01-well-A01-roi-001-exp-1_backsub.ome.tif
reading metadata
WARNING: Could not read OME metadata. Story will use generic channel names and
the scale bar will be omitted.
analyzing channel 1/38
Using marker names from markers_bs.csv
Traceback (most recent call last):
File "/app/story.py", line 159, in
main(in_path, out_path, vars(args)['m'])
File "/app/story.py", line 124, in main
vmin, vmax = auto_threshold(img)
File "/app/story.py", line 24, in auto_threshold
gmm.fit(img_log.reshape((-1,1)))
File "/usr/local/lib/python3.9/site-packages/sklearn/mixture/_base.py", line 181, in fit
self.fit_predict(X, y)
File "/usr/local/lib/python3.9/site-packages/sklearn/base.py", line 1151, in wrapper
return fit_method(estimator, *args, **kwargs)
File "/usr/local/lib/python3.9/site-packages/sklearn/mixture/_base.py", line 212, in fit_predict
X = self._validate_data(X, dtype=[np.float64, np.float32], ensure_min_samples=2)
File "/usr/local/lib/python3.9/site-packages/sklearn/base.py", line 604, in _validate_data
out = check_array(X, input_name="X", **check_params)
File "/usr/local/lib/python3.9/site-packages/sklearn/utils/validation.py", line 969, in check_array
raise ValueError(
ValueError: Found array with 0 sample(s) (shape=(0, 1)) while a minimum of 2 is required by GaussianMixture

And below is the params.yml I used:

workflow:
  start-at: background
  stop-at: downstream
  downstream: scimap
  viz: true
  segmentation: mesmer
  segmentation-channel: 1 6
  segmentation-recyze: false
  background: true
  background-method: backsub

[yujxiao]

And this is the metadata of backsub ome.tiff with --viz false flag, from ImageJ Bio-Formats plug-in:

<OME Creator="backsub v0.4.1" UUID="urn:uuid:3aa93b82-14cb-11f0-a3af-bc97e15d71b8" xmlns="http://www.openmicroscopy.org/Schemas/OME/2016-06" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openmicroscopy.org/Schemas/OME/2016-06 http://www.openmicroscopy.org/Schemas/OME/2016-06/ome.xsd">
<Image ID="Image:0" Name="rack-01-well-A01-roi-001-exp-1_backsub.ome.tif #1">
<Pixels BigEndian="false" DimensionOrder="XYCZT" ID="Pixels:0" Interleaved="false" PhysicalSizeX="0.17" PhysicalSizeXUnit="µm" PhysicalSizeY="0.17" PhysicalSizeYUnit="µm" SignificantBits="16" SizeC="38" SizeT="1" SizeX="6559" SizeY="7998" SizeZ="1" Type="uint16">
<Channel ID="Channel:0:0" Name="DAPI_1" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:0:1" Name="CD107b" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:0:2" Name="SPK1" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:0:3" Name="DAPI_2" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:0:4" Name="CAIX" SamplesPerPixel="1">
<LightPath/>
...
<TiffData IFD="0" PlaneCount="38"/>
<Image ID="Image:1" Name="rack-01-well-A01-roi-001-exp-1_backsub.ome.tif #2">
<Pixels BigEndian="false" DimensionOrder="XYCZT" ID="Pixels:1" Interleaved="false" SignificantBits="16" SizeC="38" SizeT="1" SizeX="3280" SizeY="3999" SizeZ="1" Type="uint16">
<Channel ID="Channel:1:0" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:1:1" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:1:2" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:1:3" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:1:4" SamplesPerPixel="1">
<LightPath/>
...
<MetadataOnly/>
<Image ID="Image:2" Name="rack-01-well-A01-roi-001-exp-1_backsub.ome.tif #3">
<Pixels BigEndian="false" DimensionOrder="XYCZT" ID="Pixels:2" Interleaved="false" SignificantBits="16" SizeC="38" SizeT="1" SizeX="1640" SizeY="2000" SizeZ="1" Type="uint16">
<Channel ID="Channel:2:0" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:2:1" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:2:2" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:2:3" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:2:4" SamplesPerPixel="1">
<LightPath/>
...
<MetadataOnly/>
<Image ID="Image:3" Name="rack-01-well-A01-roi-001-exp-1_backsub.ome.tif #4">
<Pixels BigEndian="false" DimensionOrder="XYCZT" ID="Pixels:3" Interleaved="false" SignificantBits="16" SizeC="38" SizeT="1" SizeX="820" SizeY="1000" SizeZ="1" Type="uint16">
<Channel ID="Channel:3:0" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:3:1" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:3:2" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:3:3" SamplesPerPixel="1">
<LightPath/>
<Channel ID="Channel:3:4" SamplesPerPixel="1">
<LightPath/>
...
<MetadataOnly/>

[ArtemSokolov]

Thanks, @yujxiao.

The issue is actually not metadata but this:

gmm.fit(img_log.reshape((-1,1)))

Auto-Minerva fits a mixture of Gaussians to approximate intensity thresholds for visualization. It looks like this fails for background-subtracted images, where the intensities have already been modified. I forwarded the issue to the Auto-Minerva developers.