[ENH] Add mrds scripts part 3 #1046
Conversation
|
Hello @arnaudbore, Thank you for updating !
Comment last updated at 2024-11-12 16:11:54 UTC |
arnaudbore
changed the title
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #1046 +/- ##
==========================================
- Coverage 68.89% 68.77% -0.13%
==========================================
Files 434 436 +2
Lines 22646 22754 +108
Branches 3078 3094 +16
==========================================
+ Hits 15603 15649 +46
- Misses 5727 5793 +66
+ Partials 1316 1312 -4
|
|
|
||
| all_metric = mrds_sum[0] | ||
| for curr_metric in mrds_sum[1:]: | ||
| all_metric += curr_metric |
There was a problem hiding this comment.
Hmm. Isn't it just all_metric= np.sum(curr_metric, axis=0)?
But it's only used for non-zero. Is there a possibility that the metrics will cancel each-other?
Something like that could be safer?:
non_zeros = np.sum(np.non_zeros(curr_metric), axis=0)
There was a problem hiding this comment.
I just add a absolute value so it does not cancel each-other, also this method mimics afd_along_streamlines.py.
| from scilpy.tractanalysis.grid_intersections import grid_intersections | ||
|
|
||
|
|
||
| def mrds_metrics_along_streamlines(sft, mrds_pdds, |
There was a problem hiding this comment.
I think that this method is not mrds-related. It's just dividing data.
It could be, in the main script:
mrds_sum, weights = \
mrds_metric_sums_along_streamlines(sft, mrds_pdds,
metrics, max_theta,
length_weighting)
weighte_mrds = weight_values(mrds_sum, weights)
Where weight values is function that divides by the weight where non-zero, somewhere in volume tools or something.
| StatefulTractogram containing the streamlines needed. | ||
| mrds_pdds : ndarray (X, Y, Z, 3*N_TENSORS) | ||
| MRDS principal diffusion directions (PDDs) of the tensors | ||
| metrics : list of ndarray (X, Y, Z, N_TENSORS) |
There was a problem hiding this comment.
I think metrics should be metric
There was a problem hiding this comment.
I aggree it's "a" list but reading it I feel like metrics make sense here.
| weight_map = np.zeros(metrics[0].shape[:-1]) | ||
| min_cos_theta = np.cos(np.radians(max_theta)) | ||
|
|
||
| all_crossed_indices = grid_intersections(sft.streamlines) |
There was a problem hiding this comment.
I am pretty sure that our "uncompress" method should be used somewhere here. @frheault probably knows this better than me.
There was a problem hiding this comment.
Maybe, this version is copy paste (kinda) of afd_along_streamlinespy
Quick description
Script to get MRDS metrics along a bundle - fixel based metric.
...
Type of change
Check the relevant options.
Provide data, screenshots, command line to test (if relevant)
...
Checklist