Float precision issue in get_confusion_matrix

[catherinetaylor2]

Describe the bug
We found that the number of true negatives was being incorrectly computed by get_confusion_matrix when we input high resolution volumes. This was due to the algorithm casting to a float:

To Reproduce
Steps to reproduce the behavior:

Expected behavior
We would expect that the sum(label + pred) <= sum(label) + sum(pred) such that the number of true negatives is not greater than the total number of voxels in the volume.