Inconsistent tissue for same tile across cycles

[josenimo]

Dear ashlar devs,

I have been optimizing ashlar for a TMA CycIF dataset, 5 cycles, 30 channels.
Now I have very good parameters that I am planning to scale to the rest of the dataset, but I have a small issue that I kind wrap my head around.. and was wondering if there is a simple solution to.
Corner tiles of my TMA cores are consistently being misregistered. I think it is because of inconsistent presence of tissue in these tiles, from what I can only assume are slide holder shifts between cycles. See picture below:

See full QC report here:
20240515_HN46_Core05_QC.pdf

Thank you for your feedback!
Happy to provide more context and qc if you think it would help

[josenimo]

Just to let you see all the cycles, and the stitching quality:

[Yu-AnChen]

Hi @josenimo, not sure if this post is under your radar - https://forum.image.sc/t/how-to-estimate-ashlar-barrel-correction-parameter/79719

We've noticed that the lens distortion is most sever on the edge of each tiles. Jeremy's script uses a simple model to correct it and after correction, the stitching and registration result are usually improved. I suspect you might need this because the absolute tile position in cycle 2 is quite different from cycle 0.