Improving metadata extraction efficiency and removing ROI Contour Sequence from metadata

[jakub-trusina]

• removed ROI Contour Sequence as those can have hundreds of MB which leads to OOM during calling js=ds.to_dict_json() .
• deletion of binary images before converting pydicom dataset (with skipped loading of large items) into json - thus we prevent reading of binary images and contours from the file

[CLAassistant]

All committers have signed the CLA.

[dmoore247]

@jakub-trusina Interesting. What's the motivation for removing this particular tag?

[jakub-trusina]

@dmoore247, it caused OOM on ds.to_json_dict(). My dataset has there hundreds of MB. When debugging, I needed 16GB cluster to process single image in Python only (not using UDF). Probably pydicom doesn't do this conversion efficiently in terms of memory usage. BTW, @erinaldidb knows the context of the project.