Alignment of Immunofluorescent and HE histological images

Dear CellProfiler forum,
My hope is that I will be able to use CellProfiler to automatically align matching sets of 2 images from a single tissue sample. The images are core biopsies that come from a breast cancer tissue microarray slide that was first stained with immunofluorescence (with an anbitody against CK8):


This stain is then washed off and the slide is stained with hematoxylin and eosin and imaged (ideally with the exact same physical parameters as the first time):


Since the images come from the exact same tissue slice, the alignment should only require rotation and translation (with no issues involving morphing of the tissue). Following alignment of the images, I would like to output 2 aligned images, which could then be used to outline the CK8 stain on the HE image. So far, I have tried a very simple CP pipeline (using a pipeline with the align module, which is attached). This pipeline does not seem to be altering the images at all (the reported shift is 0). I wonder if I am running it incorrectly, if there are other pre-processing steps I should complete prior to running align, or should I go with a different strategy altogether.

Any help is greatly appreciated!
Many thanks,
Andy Beck
Pipeline_CP_Posting.mat (882 Bytes)

Hi Andy,

The reason Align doesn’t move the images at all is that they are already aligned. They certainly look aligned to my eye, which is probably why the algorithm thinks they are aligned as well. Do you expect that they should be rotated for some reason?

~kate

Hi Kate,
I do think the images are slightly rotated. Our goal is to use the pattern of staining on the CK8 stain to label the epithelium in the HE slide. In the image below, I’ve thresholded to identify the CK8 red staining, and then overlaid the positive regions onto the HE image. The epithelial-labels are slightly rotated clockwise, so they are not correctly positioned on the epithelium:


Thanks!
–Andy

Ah, I see now. That is quite a subtle difference. My eye was not able to pick it up until you pointed it out, which is why CellProfiler couldn’t either. The align algorithm is only able to associate images based on shared areas of light and dark- and the quite varied context of H&E combined with the quite stark contrast of the cropped circle edge is probably confusing it quite a bit. What I mean is, the algorithm tries to match dark with dark and light with light- and your images have big dark areas outside of a light circle, so it thinks its done by aligning them. The very subtle staining within the tissue is lost in this. You might try masking the images first, so Cellprofiler doesn’t pay attention to the black pixels.

~kate