FIX: Housekeeping of the data loader function #92
Conversation
This PR revises the data loading: - Add the possibility to read FSL's bvec/bval files for building the gradient table - Remove reorientations of the data (still thinking why they were there in the first place) - Calculating a b=0 map if none is provided
oesteban
force-pushed
the
fix/data-loader-revision
branch
from
73aba23 to
dee4fe5
Compare
src/eddymotion/dmri.py
Outdated
| factory=lambda: Path(mkdtemp()) / "em_cache.h5", | ||
| repr=False, | ||
| ) | ||
| _filepath = attr.ib(factory=lambda: Path(mkdtemp()) / "em_cache.h5", repr=False,) |
There was a problem hiding this comment.
Looks like the black config changed?
There was a problem hiding this comment.
Yup, I can roll it back if too annoying
There was a problem hiding this comment.
NBD, so long as we don't go back and forth. At some point it will be hard to figure out what actually changed when.
| return DWI.from_filename(filename) | ||
|
|
||
| if not gradients_file: | ||
| if gradients_file: |
There was a problem hiding this comment.
Want to add some error handling here in case someone foolishly provides both kinds of input?
| b0img = nb.load(b0_file) | ||
| retval.bzero = np.asanyarray(b0img.dataobj) | ||
| elif np.any(~gradmsk): | ||
| retval.bzero = np.median(fulldata[..., ~gradmsk], axis=3) |
There was a problem hiding this comment.
I've never seen anyone use median values as the b0 image before. Is that common?
There was a problem hiding this comment.
the median is more robust against outliers, but otherwise, it should be reasonably close to the average.
There was a problem hiding this comment.
Sure. I know that. But compared to other implementations of some of these models (e.g., in DIPY) we will (sometimes? often?) get numerically different solutions, which is bound to raise questions at some point.
There was a problem hiding this comment.
The idea behind this is to have a fallback to generate a b0 quickly. The assumption is that generally, you will provide the b0_file argument with a b=0 calculated in a more sophisticated way (e.g., denoising, b1 drift compensation, etc.)
Co-authored-by: Ariel Rokem <[email protected]>
oesteban
force-pushed
the
fix/data-loader-revision
branch
from
e06de51 to
5553220
Compare
oesteban
force-pushed
the
fix/data-loader-revision
branch
from
5553220 to
7318a64
Compare
| b0img = nb.load(b0_file) | ||
| retval.bzero = np.asanyarray(b0img.dataobj) | ||
| elif np.any(~gradmsk): | ||
| retval.bzero = np.median(fulldata[..., ~gradmsk], axis=3) |
There was a problem hiding this comment.
Sure. I know that. But compared to other implementations of some of these models (e.g., in DIPY) we will (sometimes? often?) get numerically different solutions, which is bound to raise questions at some point.
src/eddymotion/dmri.py
Outdated
| factory=lambda: Path(mkdtemp()) / "em_cache.h5", | ||
| repr=False, | ||
| ) | ||
| _filepath = attr.ib(factory=lambda: Path(mkdtemp()) / "em_cache.h5", repr=False,) |
There was a problem hiding this comment.
NBD, so long as we don't go back and forth. At some point it will be hard to figure out what actually changed when.
src/eddymotion/dmri.py
Outdated
| nii = nb.Nifti1Image(data, self.affine, None) | ||
| nii.header.set_xyzt_units("mm") | ||
| nii.to_filename(filename) | ||
| return |
There was a problem hiding this comment.
Is this here for a reason?
|
Oh - sorry - just noticed that the tests are failing. Good thing they're in place 😄 |
2 participants
This PR revises the data loading: