Replies: 11 comments 12 replies
-
|
Don't know if it helps or matters for the notes, but DSP = Digital Spatial Profiler Also the 3GB png is an export of a "raw" tiff file separate from the CZI workflows |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @hansenbry - to us there doesn't seem to be one single, czi workflow. There seems to be two workflows, one which deals with large color images (which may or may not take CZI type inputs) - but if we DO take a czi input we do NOT care about having special treatments for the channels, we'll transform it as a color image into some zoomable format. The other workflow, involved with florescence - (which will take czi inputs for now), we DO care about the ability to do certain treatments with certain channels. Thoughts welcome. |
Beta Was this translation helpful? Give feedback.
-
|
The CZI files ( both H&E and FL ) will contain multiple images. Including a "label image" and a "macro image". These images are smaller and may not need a multi-scale viewer. Perhaps just a JPEG/PNG conversion? There will also be multiple "main" images and lower resolutions of them for a pyramid like format. These "main" images will need to be converted to our web based multi-scale format. Additionally, I noted that when viewing RGB images in the Zen viewer it appeared to apply to a gama of .45 to the RGB channels. |
Beta Was this translation helpful? Give feedback.
-
|
May 22 2023 What relevance have these files? All "small" files, ie: XLS / PDF / ncc / pks files should be visible to users, and available for download as a single package per collection. (Maybe - don't know if we care or not about this.) A "sample" (AKA a "collection") is a set of files in this spreadsheet. Once this is noticed by Hedwig it can be submitted by user to pipeline. The only files that the pipeline currently cares about are: the n large png file, and n czi files. (if n > 50 then throw error and don't run.) Outputs - generate thumb from "large" png file. |
Beta Was this translation helpful? Give feedback.
-
|
Will want to associate them with experiment in some way ->We will want to associate
|
Beta Was this translation helpful? Give feedback.
-
|
I checked a couple "ROI Overlay" png files of exif meta data. The tools I used did not give and additional types beyond the ones to describe the image data e.g. pixel type, resolution etc. Also for reference the "heat map" image I inspected was 2.6x2.9k size in pixels. |
Beta Was this translation helpful? Give feedback.
-
|
OK - looks like next best option would to ask that every png file to be processed have some substr prepended to file name. |
Beta Was this translation helpful? Give feedback.
-
|
Bryan and Dan are not familiar with the substr function: I suppose the other approach is to always make the smaller file "Heatmap" which I think is the default and would save me from having to append so many image files names. |
Beta Was this translation helpful? Give feedback.
-
|
A "brainstorming" idea for the heat maps which would require more work. Instead of being given a "png" for the heat map figure. The input data could be given and and interactive web visualization could be generated with python or R shiny. It could be implemented such that when a block is hovered over a pop up gives the details. https://jokergoo.github.io/InteractiveComplexHeatmap/articles/shiny_dev.html |
Beta Was this translation helpful? Give feedback.
-
|
This tool is so cool! |
Beta Was this translation helpful? Give feedback.
-
|
@hansenbry - Looking at the files that have been presented as ROI, we notice that they are all tiffs (not pngs, as suggested in the table above). We note that the version of GraphicsMagick is 1.3.38 2022-03-26 is currently being used. For For Imagemagick: |
Beta Was this translation helpful? Give feedback.
-
|
The about TIFF files are considered old, and not representative of the currently expected PNG files. |
Beta Was this translation helpful? Give feedback.
-
|
Overlay png (ROI) - do not sharpen. |
Beta Was this translation helpful? Give feedback.
-
|
Note - we'll want to create thumbs - same size as other thumbs (?) - for the heatmaps. |
Beta Was this translation helpful? Give feedback.
-
|
The steps for processing the "large png" images or ROI images consist the following steps:
|
Beta Was this translation helpful? Give feedback.
-
|
Step 1: Generate OMG-NGFF ZARR format The bioformats2raw needs to be used to convert the large png to the OME- The input png files may be RGBA ( 4-channel ) with the A ( alpha ) not containing useful information. The following is the recommended command line for the large PNG files: The compression properties were selected based to the reported performance of images of noise from this paper. . This choice follows the recommendation of the conclusion. |
Beta Was this translation helpful? Give feedback.
-
|
Step 2: Re-chunk the ZARR structure so that the RGB(A) channels are not split between chunks with the
The ZARR structure is updated in place so that the chunks are compatible for display in Neuroglancer. The resulting chunks are 780KB to 1MB size is the current estimate for an appropriate chunk size used for visualization in Neuroglancer. Note: The Alpha channel could also be removed in this step as it does not contain information ( it's all zeros ), and would reduce storage, bandwidth, and computational time. |
Beta Was this translation helpful? Give feedback.
-
|
Step 3: Create a thumbnail image. A thumbnail image can be created by extracting a level from the ZARR pyramid structure and resizing it to the desired resolution, then saving. The following Python code does this: There are a couple filtering steps done in the generations of the thumbnails to prevent aliasing. The initial plan is not to perform and sharping. And additional SimpleITK filter could be run on Step 4: Generate a key image The same as above except that the size will be different. |
Beta Was this translation helpful? Give feedback.
-
|
Step 0: Convert the input PNG to TIFF The input png uses some form of compression that slows down bioformats2raw ~100X. Converting to a uncompressed TIFF dramatically speeds up the process. The unused alpha channel can also be removed. (#249 ) The following command can be used to perform the conversion with imagemagick: |
Beta Was this translation helpful? Give feedback.
-
|
The following are the step to process large CZI files of immunofluorescence (IF) microscopy. This file consists of multiple IF images ( at multiple resolutions ) and also a "label" and "macro" color/RGB images. The IF images are multi-channel ( not visual spectrum color/RGB ) and are generally visualized with false color applied to each channel. The IF images are visualized with Neuroglancer, and no false colored thumbnails or key image, but do require neuroglacer metadata for visualization. The "label" and "macro" color/RGB images need a thumbnail and full resolution image ( no key image, as the specs for a key image are greater resolution than the original source image ). Overview |
Beta Was this translation helpful? Give feedback.
-
|
Step A: Runs ones on the original CZI file. Step A.1: Convert CZI to ZARR. Note the performance of the above conversion scale very will with the number of workers, so this is an important parameter to get reasonable performance. Step A.2: Rechunk the ZARR arrays. The ZARR structure is updated in place so that the chunks are compatible for display in Neuroglancer. This will rechunk all the extracted images in the ZARR structured. The above commands converts all images in the CZI file into one ZARR structure. In Zarr terminology the root of |
Beta Was this translation helpful? Give feedback.
-
|
Step B: Runs for each IF image in the output ZARR structure. Step B.1: Determine which ZARR sub-images are IF and multi-channel. TODO Step B.2: Generate meta-data for multi-channel visualization in Neuroglancer. Run a pytools functions called |
Beta Was this translation helpful? Give feedback.
-
|
Step C: Generate thumbnails for both the "label" and "macro" image. These images are RGB color and can be converted to standard web image file formats such as png or jpeg. Step C.1: On label image detect text orientation and rotate the image. Does the text orientation need to be detected? This would be a computational intensive task to detect the text. There are large existing libraries which do this but is this required? If we can assume that we need a 90 degree clockwise rotation of an image, then the following SimpleITK code can be used: Step C.2: Generate thumbnail Step C.3: Generate full resolutions image The same as above except that the size should be |
Beta Was this translation helpful? Give feedback.
-
Terms:
Nanostring - a device/instrument used to generate data for this workflow.
DSP - instrument
ROI region of interest
Pipepine 1: "Large 2d color images pipeline".
large pngs
large czi
IFF czi file, convert to tiff.
convert from tiff || png -> pyramidal choice.
Pipeline 2: "multi channel pipeline (2d)"
Typically light microscopy / Florescent. Possibly turn channels on off
Computationally similar to above, but will have ability to do things specific to a channel.
(Typically for now - we'll use czi.)
->
saved as png file -> very large (eg 3 gig), want to make this available to viewer.
Additionally this is saved as a CZI ("raw") file (~order of mag larger). Want to make this available to the viewer too.
Outputs:
pyramid format, viewable in a web viewer. (Maybe a thumb.)
Beta Was this translation helpful? Give feedback.
All reactions