Read single marker arrays without loading the entire acquisition into memory

[matt-sd-watson]

We have users for our IMC visualisation application who have MCDs of 10-15GB for a single ROI, which is beyond the capability of most laptops to load fully into memory. Is there API support to read individual marker arrays from MCD into memory without loading the entire acquisition? Currently, big ome tiff stacks do this much better than reading from IMC files.

[jwindhager]

IIRC, acquisition data is fully loaded to memory, including all channels, sorry. Looking at the read_acquisition source code, this should be easy to change to a memory-mapping ("on-disk") based approach, though. Basically, the culprit is here:

readimc/readimc/mcd_file.py

Line 173 in 5eb3712

img = np.zeros((num_channels, height, width), dtype=np.float32)

Perhaps the current maintainers could take a look?