bioimageanalysiscorewehi / napari_lattice Goto Github PK

Currently the interactive napari bit and batch processing are in the same library.
It would make sense to split it out and perhaps have the batch processing as an independent package so any napari imports can be avoided. For example, currently even if not using napari, we need to add the Xvfb flag in batch processing.

Batch processing is mainly here:
https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/tree/master/src/napari_lattice/cmds

The main init file contains initialisation of the logger and enums so that both batch processing and napari could use it.
https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/blob/master/src/napari_lattice/__init__.py

The parameters for batch processing are here:
https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/wiki/6.-Batch-Processing

Interactive widgets in napari:
https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/blob/master/src/napari_lattice/_dock_widget.py

Potentially use xarray instead of dask array to store data
xarray can have multiple attributes which could replace the need for setting a lot of properties in Latticedata class.
attributes can hold arbitrary metadata...

https://github.com/BiAPoL/scripts/blob/main/notebooks/xarray_example.ipynb
https://docs.xarray.dev/en/stable/user-guide/data-structures.html

Once Typer natively supports Pydantic, we may not have to duplicate any of the parameter definitions in the CLI. This will make adding new parameters much easier, as the CLI won't need to be touched. Currently this is waiting on tiangolo/typer#111 and/or tiangolo/typer#630.

Currently these defaults live in the CLI, but they should be in one consistent place, probably in the LatticeData definition.

The napari-lattice plugin has the option for batch processing within the terminal. Considering the addition of more functionalities, the number of arguments isgrowing. This can get confusing and require really long command lines which is not user-friendly. (not that running in the terminal makes it any easier 😝 )

Instead, use a yaml config file.
Syntax should be:

napari-lattice --config "location-of-config-file"

Arguments: https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/wiki/6.-Batch-Processing

When performing deconvolution and then deskewing, there is a ringing artefact on the edges.
Consider changing the padding options during decon or maybe cropping a larger area to account for this.

hi~ Can the image which captured by elyra lattice SIM (https://www.zeiss.com/microscopy/en/products/light-microscopes/super-resolution-microscopes/elyra-7.html) also use this method to process?

The end goal of this might be to be able to run mypy or pyright in the CI, to identify type errors. However the type signatures need to be improved before this can happen.

In addition to processing the image across channels and time dimensions, we may need to consider processing the image as tiles to fit large images into memory

We had thought of implementing this a while back and here is a code example:

23aa36e

Currently, cropping is only implemented for data skewed in Y direction.

Implement cropping in X direction.
This is important as future versions of Zen (>3.5 ?) will have images skewed in X direction.

Make all other functions agnostic to skew direction, which is the original goal of this plugin.

Once magicgui's guiclass is stabilized, we may be able to use this to automatically generate a GUI from the central pydantic models: https://pyapp-kit.github.io/magicgui/dataclasses/#magicgui-guiclass.

Even if this isn't the case, it will probably give us a simpler GUI implementation than magicclass.

Folks,
Thanks for making this plugin :) I have been trying to find a way to deskew/deconvolve that is not on ZEN Blue so this is great!

Here is the issue I faced - I loaded a skewed lattice light sheet dataset but napari-lattice fails to recognize the "channel" dimension in the czi file. I did try the below step as mentioned in the github page, but that did not help.

"If the channels are split into different layers, press Shift and select all the channels. Right click on the highlighted layers and then click Merge to stack."

Here is a screenshot of the error. Any ideas what the issue is? I use napari-czifile2 to read czi files in napari
Thanks!

This would allow us to decouple them from LatticeData, where they are currently stored:
https://github.com/multimeric/napari_lattice/blob/06602450b2964d16ae4be0a2cf3589ab3b8422d8/plugin/napari_lattice/_dock_widget.py#L144-L149

@haesleinhuepf showed us how we can add functions using
@register_function
juglab/napari-n2v#6 (comment)
https://github.com/clEsperanto/napari_pyclesperanto_assistant/blob/ad72ecae7d017ba63f579f05dd00be627f36e89b/napari_pyclesperanto_assistant/_napari_cle_functions.py#L7

This will enable the addition of interactive workflows.
Add options for popular algorithms or workflows not available in pyclesperanto-assistant:

• Cellpose
• ilastik
• regionprops measurements table?

Default naming of imported files is incompatible with saving at the end of the deskew. e.g. ROI_0__0 :: Image:0 :: LatticeLightsheet 3-T1

Hi all!

When I try to download napari-lattice via:
pip install git+https://github.com/BioimageAnalysisCoreWEHI/napari_lattice.git

I receive the following error:

error: subprocess-exited-with-error  
    
  × Getting requirements to build wheel did not run successfully.  
  │ exit code: 1  
  ╰─> [14 lines of output]  
      error: Multiple top-level packages discovered in a flat-layout: ['core', 'plugin', 'notebooks', 'resources', 'sample_data', 'workflow_examples'].  
        
      To avoid accidental inclusion of unwanted files or directories,
      setuptools will not proceed with this build.  
        
      If you are trying to create a single distribution with multiple packages
      on purpose, you should not rely on automatic discovery.
      Instead, consider the following options:  
        
      1. set up custom discovery ('find' directive with 'include' or 'exclude')  
      2. use a 'src-layout'  
      3. explicitly set 'py_modules' or 'packages' with a list of names  
        
      To find more information, look for "package discovery" on setuptools docs.  
      [end of output]  
    
  note: This error originates from a subprocess, and is likely not a problem with pip.

After perusing the internet I found that this error can be circumvented by either:

1. This answer on stackoverflow suggested adding the following lines to the pyproject.toml:

[tool.setuptools]  
py-modules = []

2. Since I don't have access to the repo, I tried downgrading setuptools via:
   pip install setuptools==version#

Unfortunately downgrading setuptools still causes a host of errors to occur. Any suggestions?

Thank you so much !!!

Using RedLionFish for deconvolution

rosalindfranklininstitute/RedLionfish#12

Refer to Volkers notebook, where he uses thresholds to figure out bounds to crop the PSF:

https://github.com/VolkerH/Lattice_Lightsheet_Deskew_Deconv/blob/master/examples/find_PSF_support.ipynb

Use this for cropping the experimental PSF or deconvolution using RedLionFish can fail due to memory issues.

Keep data in GPU memory if deskewing

I'm envisaging pre-commit + black + isort so we don't have to worry about code formatting.

Currently, napari-lattice works only with a single ROI and it assumes that the cell doesn't move much.
If cell wobbles a bit, we just draw a bigger ROI.
This is all done using the MIP image after image acquisition for Zeiss lattice.

We need to define how this is going to look like, i..e., scheme of the pydantic model if using this approach and best way to integrate it.

Currently, napari-lattice Workflow module iterates through each channel and applies the custom processing defined in the workflow.
For cellpose, if a multichannel image is given, this can be used as input for improved prediction using cyto models.
For example if channel 1 is nuclei and channel 2 is cytoplasm, segmentation maybe more accurate on the combined multichannel image than on channel 1 or two individually.

Either new function for this or alter the code here:

• Swap "Cropping" and "Deskewing" sections
• Add more details on installing napari, perhaps point to full documentation?

Hi @haesleinhuepf
I'm testing out napari_lattice and learning how to publish it as a pypi package.
I'm having trouble setting up github actions and automated testing/deployment.
Would you mind helping me out with this?

Cheers

Pradeep

https://forum.image.sc/t/richardson-lucy-deconvolution-large-images/85325/10?u=pr4deepr
https://github.com/clij/clij2-fft/tree/master

Uses clij and OpenCL for deconvolution.

Use information from this page to customize the listing on napari hub as well already have the actions running

https://github.com/chanzuckerberg/napari-hub/wiki/Customizing-your-plugin's-listing

We assume that the pixel size of the psfs are same as that of the image and use that throughout the code.
Its worth adding this note in the documentation and ask the user to verify this is the case, i.e., psf should be acquired using same settings as that of the image

• Define a class for writer so its easier to incorporate multiple formats, initialize and save images. Add flexibility to save images of any dimension (currently only 3d images can be changed..
• Security issue: Currently workflows option runs .py files in a folder specified by user. This is a potential security issue. Add option so user can confirm which files to run.
• Define a function to handle the return types from workflows. For example, if multiple object are returned, they need to be in a tuple. Otherwise, treat it as a single object.
• use plugin_register to add different software support so workflows can be initialised from napari: #17

Currently, h5 format refers to bigdataviewer/bigstitcher compatible h5.
Need to verify how compatible and versatile this h5 format is with other software.

• Save as traditional h5: follow this notebook for reference: https://github.com/ilastik/ilastik/blob/main/notebooks/h5convert/convert_to_h5.ipynb
• Save as bdv h5
• Open both files in software like ilastik, napari and Fiji.

As we are using a single affine transformation for shearing and rotation, this can introduce artefacts in the final data, especially on non-isotropic data. @haesleinhuepf has raised it here:
clEsperanto/pyclesperanto_prototype#196

Turns out, this is a well-known issue and has been documented here:
VolkerH/Lattice_Lightsheet_Deskew_Deconv#22
napari/napari#1843 @haesleinhuepf , @VolkerH has discussed an OpenCL-related solution/fix? Not sure I fully understand it

We could technically chain the affine transformations individually, but the intermediate sheared image will be extremely large and cause memory issues on the GPU.
Another option is to write an Opencl kernel specifically for deskewing. This code from OPM can be used for inspiration:
https://github.com/QI2lab/OPM/blob/5dc5a4f2046d220e09d038ae6d292f3590e4f015/reconstruction/image_post_processing.py#L33

napari supports setting the affine transforms and viewing the transformed 3D volume:

https://napari.org/dev/gallery/affine_transforms.html

https://forum.image.sc/t/napari-performance-on-a-tb-czi-image/84587/21?u=pr4deepr

Have a checkbox for live deskew and use this instead of the 'Preview Option'. Ensure the ROI coordinates match and generating a MIP is fast

the pip bug that was causing installation issues has been fixed as of pip 23.3, so if you are interested we could fix all the requirements in this repo and just ask users to update their pip?

The time end entry option is limited to 999.

If a user has configured the GUI and happy with the settings, batch processing using the CLI would be the next step.
To make the transition easier, add option to export the schema in the form of a yaml config file that can be ingested by CLI.

• Add more documentation on using workflows
• How to use napari-assistant, export workflow and apply it on a lattice dataset
• How to run custom python scripts
• Use ilastik classifier (limiting feature sets in ilastik to tackle memory issues)
• Example workflows on datasets for segmentation and measuring morphological properties

Ideally a library will have its own custom error hierarchy so a user can except those errors specifically. It might be useful to therefore have LlsError and some subclasses of that. This is low priority however.