macsima reader fails on experimental data due to inconsistent image dimensions

[nbonine]

The spatialdata_io.macsima reader currently assumes that all TIFF files within a MACSima ROI folder have identical pixel dimensions. While this holds true for demo datasets, actual experimental MACSima data frequently contains images with slight variations in height or width, even after standard pre-processing. This causes the reader to raise ValueError: Images are not all the same size.
@MSHelm

[nbonine]

The harpy library already addresses this shape-mismatch in their MACSima reader (see https://github.com/saeyslab/harpy/blob/main/src/harpy/io/_macsima.py). However, the implementation would benefit from some slight adjustments, as discussed in saeyslab/harpy#187.

[MSHelm]

Hi @nbonine thanks for the report. To be honest I havent encountered data sets that have a shape mismatch, but I also just started using spatialdata for analysis, so it might have gone unnoticed so far. Do you have a dataset that shows that mismatch that you can share, so I could have a look into it?

[MSHelm]

Also, do you use the preprocessing that is done during acquisition, or do you run a dedicated preprocessing step after acquisition has finished?

[nbonine]

Unfortunately, I do not have data I can share publicly. As a Miltenyi employee, I assume you have access to many more MACSima datasets than I do.
For preprocessing, the default preprocessing by MACSiQ was used.

[MSHelm]

Hi @nbonine
I ran the current macsima reader version on several internal data sets and couldnt reproduce your error. All of them have identical X/Y dimensions.

When you say you did default preprocessing, do you mean by that the preprocessing done on the machine during acquisition? We recommend to run a dedicate offline preprocessing step afterwards, as then all data is available and it is more robust. Could you try if this solves your problem?

If the issue persists, I would need some kind of data to work with. Maybe you can send some single DAPI images or so, that dont contain too sensitive information, or you could randomize the image content. Alternatively maybe going via our Tech Support could be an option if you think sharing with them is a more viable option for you.

As a side note, the demo data sets I reference in the reader are not created differently than any other MACSima experiment ;)

[nbonine]

@MSHelm I've sent you an email with some example data!