New script to compute similarity between mask and atlas (nii) #994
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #994 +/- ##
==========================================
- Coverage 68.61% 68.31% -0.30%
==========================================
Files 426 427 +1
Lines 21834 21940 +106
Branches 3267 3286 +19
==========================================
+ Hits 14981 14989 +8
- Misses 5578 5676 +98
Partials 1275 1275
|
arnaudbore
requested review from
arnaudbore,
AntoineTheb,
CHrlS98 and
gagnonanthony
| img_1, dtype_1 = load_img(filename_1) | ||
|
|
||
| if np.issubdtype(dtype_1, np.floating): | ||
| raise ValueError('Input file {} is of type float.'.format(filename_1)) |
There was a problem hiding this comment.
Why only float ? The comments in the header of the script mention that the inputs must be uint8 or uint16, but here int32 would pass the test.
There was a problem hiding this comment.
This was simply to make the error message a bit clearer in the context of this script (if used from Freesurfer atlas for example)
The function get_data_as_labels() will convert int32 to uint16 automatically
| binary_2[data_2 == val] = 1 | ||
|
|
||
| # These measures are in mm^3 | ||
| volume_overlap = np.count_nonzero(binary_1 * binary_2) |
There was a problem hiding this comment.
See if you could use:
scilpy.tractanalysis.scoring.compute_f1_overlap_overreach, which can be tested with unit tests.
There was a problem hiding this comment.
Since this does not always use , it is not always normalized between 0 - 1 (this script is not for GT). So I will leave it like that.
|
Hello @frheault, Thank you for updating !
Comment last updated at 2024-08-05 18:09:41 UTC |
|
@ThoumyreStanislas Could you try this branch before merging to compare atlas (labels of bundle in scan-rescan, I think you have penthera) and maybe lesions masks? That would be helpful to improve the docstring |
There was a problem hiding this comment.
script works just as well for masks as for bundle labels, the only thing is that json is not the easiest to understand. In particular, when comparing binary masks with the “1” section in the json and the order of inputs in each section, even if we can guess it, it might be good to explain exactly how the classification works in the json.
Quick description
This is similar to scil_bundle_pairwise_comparison.py, but for mask and atlas (as nifiti). This is an easy way to quantify the similarity between wmparc of bundle label maps or BET mask in scan-rescan or across subjects for example.
This is generating a JSON file that contains four measures: dice, adjacency_voxels, overlap and overreach.
Everything has to be in uint8 or uint16, coregistered nifti.
...
Type of change
Check the relevant options.
Provide data, screenshots, command line to test (if relevant)
...
Checklist