Support multimask segmentations

[Zach Pearson]

As reported by Selina, SEG models in NSCLC-Radiomics-Interobserver1 have "multimask" segmentations. These are X x Y x ((num_segmentations) x Z) sized files where every Z frames a new segmentation starts.

[Zach Pearson]

The series class separates seriesfiles out by size already, and creates a DicomData class for each detected size of images. This was done to support PET scans that have differently sized images in the same series.

[Zach Pearson]

So maybe when we create a series file we can interrogate it further and if it's got more than one segmentation in it we create a DicomData instance for each one. Alternatively, we can interrogate it during open_grids to see if it has a number of frames that's an integer multiple of the number of frames in the image it references.

[Zach Pearson]

I was able to detect the multimask segmentation and open them separately. The issue now is that they don't have the same SliceSpacing as the scan they refer to. Maybe the SliceSpacing is proportional to how many masks there are.

[Zach Pearson]

Added some code to the SeriesFile class to try and detect multimask segmentations, tested on 1.3.6.1.4.1.9590.100.1.2.103517357412192184039370205801611698844

If the segmentation is multimask then we set the Z spacing based on the mask length instead of the number of files.

Looks like other datasets are still working after the additions, which are ​here.