Error with registration when I use a cropped version of the image

[msberges]

Normally I am working with images that have the same cellular tissue two or more times. The problem I have is that the medical notes are only in one of them.

Sometimes when I do the registration it is registered with another of the cellular tissues where there are no annotations, so when I transfer them from one image to another it does not do it well.

For this reason, one way to solve it is to crop the original image with only the cellular tissue of interest (a roi).

(I know that some annotations are missing in this example, it is for another reason and it should not affect the register)

But when doing this the normal registration has stopped working in most cases.

I do not understand what is happening. When there were two cellular tissues, the record was done well but just with the one that had no annotations, but when there is only one, the record goes terribly wrong.

Anyone know what might be happening? Could cropping the image have affected the registration in any way?

[cdgatenbee]

Hi @msberges,

Sorry you're running into this issue, but could you please clarify a few things for me?

• When you say "Sometimes when I do the registration it is registered with another of the cellular tissues where there are no annotations, so when I transfer them from one image to another it does not do it well" is the problem that the alignment works, but the annotations aren't warping correctly, and so there is a bug in valis that needs fixing? Or, are the annotations being warped correctly, but the issue is that sometimes there is no tissue in the area where the annotations are transferred?

• It looks like something odd is going on the image associated with the green overlay. The processed version looks very blocky, but I would have thought it would more like the other image (the pink overlay, associated with annotations?). Is the cropped image the one associated green overlay, and if so, was it saved at the same resolution as the original?

• The thumbnails in the "masks" folder should show how valis is converting the full resolution image to the lower resolution one used for registration, with an outline drawn around the regions to be registered. Do these images look as expected? This sort of gets back to whether or not something is going on during the downsampling and/or pre-processing valis does prior to aligning the images.

I think this extra info will help narrow down what's going wrong here.

Best, -Chandler

[msberges]

What I mean is that, as seen in the first image, I have many mrxs files that when opened with qupath have the same tissue repeated 2, 3 or 4 times (it is the same cellular tissue, but not exactly the same photo) . This only happens to me with the HE stain, not with the HER2, KI67, etc. stains.

Another example: HE stain: RE stain:

As can also be seen in the image of the first comment, the annotations are only in one of the cellular tissues (in this case in the one below).

If I ask the program to register the images, sometimes the image in HER2, KI67... will overlap with the tissue above, the one that has no annotations. Therefore, when transferring them, the program will do it wrong.

For this reason I crop the images to only keep the cellular tissue that has the annotations. The problem is that by having one image cropped and the other not, the low quality versions used in the registry have the tissue area with a very different resolution. That's why in processed images, the uncropped image looks so blocky compared to the cropped one.

The solution I found for this problem, to always ensure a good register, is to crop the two ROIs in both images and adjust the annotations at each step. Here an example of hat I did: From the original source image to the cropped source image: From the cropped source image to the cropped target image: From the cropped target image to the original target image:

This is the solution that I have found and it works very well, although it is probably not the most efficient.