Fiji ilastik Bridge
Updated 1,531 Days AgoPublic
Actions

This page explains a way to use ilastik classification from FIJI/ImageJ, using the BIOP FIJI ilastik Bridge. Another parallel is the official ilastik FIJI plugin (https://www.ilastik.org/documentation/fiji_export/plugin). The BIOP plugin has the advantage that it does not require conversion of files to hdf5.

NOTE: You cannot do the training within FIJI, this needs to be done in ilastik. However, once you have a properly set ilastik project file (with .ilp extension), you'll be able to perform ilastik classification from within FIJI/ImageJ, in a convenient and easily scriptable/recordable manner.

1 - Setting up the FIJI Ilastik Bridge ( needs to be done once )

1.a. Install ilastik on your computer

Installer can be find on https://www.ilastik.org/

1.b. Enable BIOP fiji udate dev site:

In FIJI, click → Update → Add Update Site. In the new line created:

put any name you like in the column Name (like BIOP-DEV)
put https://biop.epfl.ch/Fiji-Update-Dev/ in the URL column
click Close then Apply changes, then restart FIJI

1.c. Make the connection between ilastik and Fiji

Fiji needs to know where the main ilastik executable file is located.
This needs to be done once. After, FIJI will remember it (as long as you do not remove or change ilastik location on your HDD)

In FIJI : click Plugins → BIOP → Ilastik → Set Ilastik Location

Click Browse in front of IlastikExecutable (take care : do not choose Elastix, it's something different! ) and select the location of the ilastik.exe file.
- Typically on windows: C:\Program Files\ilastik-[version]\ilastik.exe
- TODO : find typical ilastik location in Mac
Press OK and then click Window > Console.

You should see:

[INFO] Ilastik was found ;-)

2 - Practical example : cell / nuclei pixel classification

2.a. Make a pixel classifier with ilastik

Use ilastik as usual to make your classifier.

WARNING: Set the appropriate export settings

It's important to export the files in multipage tiff, please follow the settings below:

2.b. Use Ilastik classification within FIJI

With a FIJI properly set up:

Open the image you'd like to classify.
Click Plugins → BIOP → Ilastik → Ilastik Pixel Classification (or use the search bar to directly find the Ilastik Pixel Classification command).

Choose the right settings:
- Ilastik Project File : select the ilp file you'd like the use for the classification (pixel classifier with proper export settings).
- Classifier output :
  - Simple Segmentation : outputs a label image containing, for each pixel, the value of the associated class.
  - Probabilities : returns the probability map for each class (one channel per class). Use float32 as pixel type for a nice output.
  - Pixel type :
    - uint8, for label image that contains less than 255 labels (typical for pixel classifier).
    - uint16, for label images that can contain more than 255 labels and up to 65535 (could be useful for object classification and big images).
    - float32, for probability maps.

Click OK
Wait for the classifier to have finished its job. A result image should pop up.

NOTE: If you see a white output window, try changing the LUT. For instance Image>Lookup Tables>glasbey

These commands are Macro Recordable!

3 - More information about Ilastik FIJI Bridge

3.a. What's currently supported in the ilastik FIJI Bridge

What is supported by the bridge:

Image type supported:
- Single fluorescence channel
- RGB Images
Ilastik taks:
- Pixel classification
- Multicut
Export:
- Label image
- Probability map

What's known as currently unsupported:

Image type supported:
- Multi fluorescence images
- Time Series ( to check )
Ilastik taks:
- Object classification only

3.b. Source Code

Hosted on the BIOP GitHub account : https://github.com/BIOP/ijl-utilities-wrappers

4 - Troubleshooting

If you get an error like java.lang.UnsupportedOperationException: Attempting to read or write from a io.scif.io.VirtualHandle, this probably means that ilastik has failed for whatever reason

Make sure that you select an appropriate image before starting the classifier:
- Make sure that the original image is selected, not the results from a previous classification
- Make sure it is the correct bit depth and correct number of channels)

(ERROR) Feel free to report to the BIOP in case the error cannot be solved (https://biop.epfl.ch)

If you do not see any result images appearing after a very long time, maybe an error has occured. Check the error message in the console window. You can also restart the task and click verbose to get more information about the error.