Full FOV two-photon image QC

[nataliaorlova]

Full FOV 2P images are used for ophys-omfish registration and are going to start being generated by LIMS mesoscopes splitter module starting the week of Feb 13. We should think about how to QC them.

Example image can be found here

The file will be located at $SESSION_OUTPUT_DIR/${session_id}_stitched_full_field_img.h5. In the h5 file one can find a stitched full FOV 2P image and stitched full FOV 2P image with averaged surfaces inserted into it, along with the raw tiff metadata.

Once the images are available we should review a few examples and discuss features/metrics to QC. First thoughts are:

• Full FOV 2p image is stitched correctly
• Surfaces are inserted
• Brightness/contrast is of appropriate values (not too dark, not too bright, offsets correct)

[nataliaorlova]

Here is the first file example generated by LIMS:

"\allen\programs\mindscope\production\vipssn\prod0\specimen_1210632453\ophys_session_1247774713\1247774713_stitched_full_field_img.h5"

[nataliaorlova]

Will catch up with @cbrrrry too spec out the skeleton of what QCing this image should be.

[nataliaorlova]

Carson (@cbrrrry ) and I talked yesterday and here is a skeleton of the QC for this image/file:

1. hdf5 file (examples name 1268323260_stitched_full_field_img.h5) is present and has the following fields:
   • 'full_field_metadata' (SI metadata dictionary for the full field stack)
   • 'stitched_full_field', (stitched full field image)
   • 'stitched_full_field_with_rois', (stitched full field image with mapped into it ROIs)
   • 'surface_roi_metadata' (SI metadata for surface avg images)
   • 'insert_coordinates' (dictionary with coordinates to insert each ROI into Full Field image )
2. Both stitched mapped and just stitched image is presenter to QCer to check:
   • whether there are no saturated features in the FOV and vasculature is visible - if there is saturation, flag for needing re-normalization or cropping.
   • stitching is correct in 'stitched_full_field' image: there is no significant (>10 pixels) horizontal overlap or vertical misalignment. If there is - fail and flag needing Natalia's re-calibration.
   • entire cranial window is captured in 'stitched_full_field' image.
   • all ROIs are present in the stitched_full_field_with_rois image and their location is correct (with 50 pixels accuracy)
   • FOV cleanliness: no bright speckles on the surface of the window.
   • note if bubbles are obviously visible (dark round or oval hazy spots)

[DowntonCrabby]

@nataliaorlova can you provide code that does 1?

For 2- if these items are evaluated manually (ie the operator looks by eye) can you please provide example images of the flag and failure modalities that you described above? I think we already have a controlled language tag and image for the bubbles but everything else is new

[nataliaorlova]

The code for finding the file, opening it and checking the fields? You guys should be able to do that in 5-10 lines of code...

As to #2 - let's start by visualizing the image to the operators - this is a new data stream, and we will be collecting examples as we have a way to look at them.

[cbrrrry]

I have some rough code for this, part of which you have previously shared @nataliaorlova, that I am happy to provide. Would this live in data_access @DowntonCrabby ?

[DowntonCrabby]

Okay had a brief meeting and here were the take-aways & tasks:

Proof of concept coding: Carson to provide code that does these tasks:

• [x] check for the existence of the stitched_full_field_img.h5 & the 5 fields
• [x] if exists save images as matplotlib figures

qc metric generation pipeline:

• [ ] Sean &/or Clark to provide location in brain_observatory_qc to put the code
• [ ] Sean, Arielle or Clark to implement the code in mindscope_qc_metrics to make it part of the pipeline

qc module creation & display in report

• [ ] Clark to add new image metrics in mongodb
• [ ] Clark to create new controlled language tags based on the flag/failure modalities outlined above by Natalia
• [ ] Clark to create new report module to display these metrics

[DowntonCrabby]

Just for visibility/clarity- we will try to incorporate this into the next milestone/release.

We are currently in the phase doing bug fixes on Milestone 2- so we're talking about this being in like milestone 3

[cbrrrry]

I submitted a PR called " Fullfield qc #292 " with the code for this issue. Review at your leisure!

[jkim0731]

There are some cases where stitching was not good: E.g.: \allen\programs\mindscope\production\learning\prod0\specimen_1285648947\ophys_session_1305964574

[jkim0731]

Sometimes the fullfield is not stitched E.g.: \allen\programs\mindscope\production\learning\prod0\specimen_1292035016\ophys_session_1300393608

[matchings]

@DowntonCrabby @seanmcculloch I think this is still needed and should be prioritized as part of mouse QC / CO migration work