RF: Anatomical processing #286
Conversation
- Precomputed aseg should be connected via the inputnode, no need to validate the image again. - MCRIBS code will be moved to the MCRIBS-specific surface workflow
Codecov ReportPatch coverage:
❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more. Additional details and impacted files@@ Coverage Diff @@
## master #286 +/- ##
==========================================
+ Coverage 32.07% 32.18% +0.11%
==========================================
Files 53 54 +1
Lines 4727 4748 +21
==========================================
+ Hits 1516 1528 +12
- Misses 3211 3220 +9
☔ View full report in Codecov by Sentry. |
| return os.path.join(base, basename) | ||
|
|
||
|
|
||
| def get_file(pkg: str, src_path: Union[str, Path]) -> str: |
There was a problem hiding this comment.
I like this. But I would consider using a module-local ExitStack and cached responses to avoid unnecessary duplicates.
Example:
import atexit
from contextlib import ExitStack
from functools import cache
try:
from importlib.resources import as_file, files
except ImportError:
from importlib_resources import as_file, files
# Module-scoped
exit_stack = ExitStack()
atexit.register(exit_stack.close)
@cache
def get_file(pkg: str, src_path: Union[str, Path]) -> str:
file_ref = files(pkg) / src_path
pathlike = exit_stack.enter_context(as_file(file_ref))
return str(pathlike)| from nipype.interfaces import utility as niu | ||
| from nipype.pipeline import engine as pe | ||
| from niworkflows.engine.workflows import LiterateWorkflow | ||
| from niworkflows.utils.spaces import Reference, SpatialReferences |
There was a problem hiding this comment.
This can be an expensive import. I might use a from __future__ import annotations to avoid runtime resolution of annotations, and then put this in an if ty.TYPE_CHECKING: block.
| t1w_preproc_wf = init_anat_preproc_wf(name="t1w_preproc_wf") | ||
| t2w_preproc_wf = init_anat_preproc_wf(name="t2w_preproc_wf") |
There was a problem hiding this comment.
Are we making strong assumptions that num_t1w and num_t2w >0? Even if it's true for now, I would consider guarding these.
The anatomical workflow was starting to become a rats nest, with a variety of workflows that may or may not be called depending on the presence of precomputed derivatives (
--derivatives). This is an attempt to untangle the workflow into clearer steps. A top-level breakdown of the major reorganization this PR addresses:Template workflows (T1w / T2w)
Create a canonically-oriented, conformed reference image. If multiple runs are found, intensity-nonuniform (INU) correction is run on a each volume, and are used to create a single volume structural reference. Additionally, if a precomputed mask / aseg are provided (only if a single T1w is found), they are reoriented and conformed as well.
Preproc workflows (T1w/T2w)
Clean up anatomical reference by performing intensity clipping, denoising, and INU correction.
Brain Extraction workflow (T2w)
If a precomputed mask is not available, register the T2w (higher contrast) to a skull-stripping template to create a brainmask.
Coregistration workflow (T2w -> T1w)
Align the T2w to T1w space, and optionally the mask (if the brain extraction workflow was needed).
No remarkable changes to the normalization, segmentation, and surface generation workflows.