Cannot export pixel segmentation image from ilastik 1.3.2

I am trying to convert bright field images into fluorescent images using ilastik 1.3.2. My source image is RGB color 960 x 1280, 3, uint8 tiff format. The pixel segmentation pipeline works well to clearly separate cytoplasm, nuclei, biomarker (CtBP) and background with four labels. But when I export single segmentation image with tif sequence format, there is an error message: Wrong number of non-channel axes for format tiff. If I click “transpose to axes order” to change from yxc to cyx, there is another error message: Failed to generate export file: ……Exported stacks must have exactly 3 non-singleton dimentions (other than the chanel dimention). Your stack dimentions are: OrderDict([(‘c’,1),(‘y’,960),(‘x’,1280)]). I have no idea what’s wrong with it. Is it a source image format problem or export options setting issue or another reason? Any advice to fix the problem is very appreciated

(3.5 MB)

Hi @yhsu822,

I should have checked here first before replying you via email.

So in short - what you have discovered might be a bug that should be fixed in version 1.3.3post2.
As an alternative you can also export to .h5 with the default settings, and then import the image to ilastik using our fiji import/export plugin.

Cheers!

Hi Dominik,
I really appreciate your quick reply. I installed 1.3.3post2 version but the export problem is the same as before. I do can export .h5 format image with default export options. Then I can import ,h5 image into ImageJ through ilastik plugin. I am confused now how to process the image next in order that I can use CellProfiler to do high throughput image analysis. I tried to convert .h5 into .tiff format but the color image did not have multiple channels. So I have no way to split channels for downstream analysis. Is there a better way to convert bright field image (RGB color 960 x 1280, 3, uint8 tiff format) into fluorescent image with same tiff format? I ever used ilastik without export problem and following CellProfiler analysis did very well. Any more advices or any suggestions so that I may try? Thanks!

One question: After I installed ilastik 1.3.3post2, shall I uninstall previous versions like ilastik 1.3.2 and 1.3.2post1?

Hi @Tim_Su, you can have multiple versions of ilastik - but I would still remove the older one(s) if there is no obvious reason to keep using those.

Hi @Tim_Su,

regarding your problem. Right, in order to use ilastik in cell profiler you’ll have to export it to something cellprofiler can read. Luckily there is an ilastik cell profiler plugin (that goes by the very explicit name “predict”) here: https://github.com/CellProfiler/CellProfiler-plugins that should take care of the data export for you.

Maybe try it out. I haven’t ever used it myself, but found something in my notes mentioning that one has to add a “colortogray” module after the “predict” one.

For your original question: The number of channels in your input image does not influence the number of channels in your output image. The number of channels in your output image depends on the number of Classes/Labels you have added.

Sounds great. I will try ilastik cell profiler plugin and let your know later. Thanks so much Dominik!!