Accuracy of identity matrix assumption

[DanNBullock]

validator-neuro-anat/run.py

Lines 48 to 57 in b7c5142

	def check_affine(affine):
	if affine[0][0] != 1: results['warnings'].append("transform matrix 0.1 is not 1")
	if affine[0][1] != 0: results['warnings'].append("transform matrix 0.2 is not 0")
	if affine[0][2] != 0: results['warnings'].append("transform matrix 0.2 is not 0")
	if affine[1][0] != 0: results['warnings'].append("transform matrix 1.0 is not 0")
	if affine[1][1] != 1: results['warnings'].append("transform matrix 1.1 is not 1")
	if affine[1][2] != 0: results['warnings'].append("transform matrix 1.2 is non 0")
	if affine[2][0] != 0: results['warnings'].append("transform matrix 2.0 is not 0")
	if affine[2][1] != 0: results['warnings'].append("transform matrix 2.1 is not 0")
	if affine[2][2] != 1: results['warnings'].append("transform matrix 2.2 is not 1")

I could certainly be wrong about this, but I don't think it is the case that T1 images necessarily have an identity matrix here. For example, it could be (for whatever reason) slightly non-orthogonal, and thus have non-zero entries on the off diagonal. Perhaps this is undesirable, and that's what the warning is trying to nudge about? Even so, I believe that it is quite possible for T1 images to have sub-millimeter resolutions (i.e. .75), and it would be strange to nudge them (via warning) about a supererogatory characteristic.

[soichih]

I agree and I always thought it's odd that we have to make this assumption, but I was told in the past that without affine being ident subsequent processing might fail. I think LAS vs RAS is also the same type of assumption. Ideally, all of our apps should work equally well with data being either LAS or RAS..

This is a very old assumption, so if we know that all of our App works well with non ident affine, then we can just drop this warning.