Displaying Multi-channel images and labels in Interactive Segmentation View

Hi -

We are exploring the possibility of using a pure KNIME+KNIP (without ImageJ macros, etc.) based interactive workflow to manually select segmented objects (labels) from multichannel (4 spectral channels) fluorescent images.

The KNIME workflow as well a sample multi-channel (made-up) are attached. The issue we are facing is more to do with the inability of the “Interactive Segmentation View” node to display multi-channel image in a pseudocolor along with segmented objects (in bounding box mode). That is, is there a way to display multi-channel images in a fused color (composite) mode along with ROIs similar to Fiji? For your convenience, I have also included ImageJ ROIs (.zip) for the example input image so you can import them in Fiji and overlay on the attached multi-channel (composite mode) input image.

The use case is as follows:
Segment/label objects from different channels --> First, select an object from one channel (e.g., Segment 1 from Ch2) --> Annotate it (Interactive HiLite Collector) --> Then, select corresponding object(s) from another channel (e.g., Segments 1 and 2 from Ch1) that are proximal to previously selected/annotated object (Segment1 from Ch2).

Constant swapping of “Channels” in the “Interactive Segmentation View” node can be visually tiresome, especially if the image has several regions (e.g., nuclei/cells) from which the user has to make these sort of selections/annotations.

Any thoughts/suggestions/comments?

Thanks,

Reddy

Attachments:

multichannel_input.tif (326.8 KB)
multichannel_input_rois.zip (626 Bytes)
multichannel_interactive_segmentation_view.knwf (40.3 KB)

2 Likes

Dear @reddyg,

This is currently not implemented and I would agree that the Interactive Segmentation View should allow you to change the representation of the underlying image and the labeling independently (as you can do with a multi-selection in an Image Viewer). There is a similar issue concerning the Interactive Annotator that has the same limitation.

I am not entirely sure that I understand the need for switching channels in the attached workflow. Would converting your multichannel image to a PNG be a viable workaround for switching the channels?

Anyhow, the ColorTable based Image Renderer doesn’t seem to allow for switching to a composite view of the image (in contrast to the RGB Image Renderer which does this automatically for images with Channels ≤ 3). I have created https://github.com/knime-ip/knip/issues/519 to keep track of this.

Thanks for reporting, @reddyg!

Best,
Stefan

2 Likes