Analyze midsagittal lesion length and width #96

valosekj · 2024-09-20T16:40:25Z

This PR adds a codebase to compare the midsagittal lesion length and width (recently added in spinalcordtoolbox/spinalcordtoolbox#4617) computed using different methods:

manually measured metrics
metrics computed using sct_analyze_lesion from GT masks (located under derivatives/labels)
metrics computed using sct_analyze_lesion from the lesion and SC masks obtained using SCIsegV2
metrics computed using sct_analyze_lesion from the lesion obtained using SCIsegV2 and SC obtained using the contrast-agnostic model

Note: The contrast-agnostic model is tried to see how it performs compared to the SCIsegV2 model. Comments below show some screenshots of SC seg obtained using the contrast-agnostic model.

At the same time, this PR aims to investigate an alternative method for getting the midsagittal slice (proposed in spinalcordtoolbox/spinalcordtoolbox#4631).

For details, visit this README.

… compute the midsagittal lesion length and width.

… we use '-ofolder' for sct_analyze_lesion, then QC report fails.

…e previous commit

…ic model for now Context: spinalcordtoolbox/spinalcordtoolbox#4633

valosekj · 2024-09-20T16:44:10Z

RESULTS FROM contrast-agnostic model (version: 2.4)

Adding -largest 1 for sct_deepseg in 8d9b7b9 as the contrast agnostic model segments areas outside of the cord for some subjects (especially lumbar cord with metal implants):

https://github.com/ivadomed/model_seg_sci/pull/96\#issuecomment-2364121461

valosekj · 2024-09-20T16:51:09Z

RESULTS FROM contrast-agnostic model (version: 2.4)

When calling sct_analyze_lesion on the spinal cord obtained by the contrast-agnostic model v2.4 and lesion segmentation obtained by SCIsegV2:

model_seg_sci/midsagittal_measures/01_compute_midsagittal_lesion_length_and_width.sh

Lines 127 to 128 in 8d9b7b9

    
           # Compute the midsagittal lesion length and width based on the spinal cord obtained using the contrast-agnostic model v2.4 and lesion segmentation obtained using SCIsegV2 
        
           sct_analyze_lesion -m ${file_t2}_lesion_seg.nii.gz -s ${file_t2}_sc_seg_contrast-agnostic.nii.gz -qc ${PATH_QC} -qc-subject ${SUBJECT}

sct_analyze_lesion fails because SC seg might not cover the lesion:

full error output

+ sct_analyze_lesion -m sub-zh04_ses-02_acq-sag_T2w_lesion_seg.nii.gz -s sub-zh04_ses-02_acq-sag_T2w_sc_seg_contrast-agnostic.nii.gz -qc /home/GRAMES.POLYMTL.CA/p118175/results/sci-zurich_midsagittal_measures_2024-09-20/qc -qc-subject sub-zh04/ses-02

--
Spinal Cord Toolbox (git-master-197ef3494a1c73d8b8acb14fcd809f31e16b5265)

sct_analyze_lesion -m sub-zh04_ses-02_acq-sag_T2w_lesion_seg.nii.gz -s sub-zh04_ses-02_acq-sag_T2w_sc_seg_contrast-agnostic.nii.gz -qc /home/GRAMES.POLYMTL.CA/p118175/results/sci-zurich_midsagittal_measures_2024-09-20/qc -qc-subject sub-zh04/ses-02
--

Creating temporary folder (/tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau)
cp sub-zh04_ses-02_acq-sag_T2w_lesion_seg.nii.gz /tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau
cp sub-zh04_ses-02_acq-sag_T2w_sc_seg_contrast-agnostic.nii.gz /tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau

Orient input image(s) to RPI orientation...
^[[33mFile sub-zh04_ses-02_acq-sag_T2w_lesion_seg.nii.gz already exists. Will overwrite it.^[[0m
^[[33mFile sub-zh04_ses-02_acq-sag_T2w_sc_seg_contrast-agnostic.nii.gz already exists. Will overwrite it.^[[0m

Label connected regions of the masked image...
^[[33mImage header specifies datatype 'uint8', but array is of type 'int64'. Header metadata will be overwritten to use 'int64'.^[[0m
/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/spinalcordtoolbox/scripts/sct_analyze_lesion.py:394: RuntimeWarning: divide by zero encountered in scalar divide
  axial_damage_ratio_dict[slice] = lesion_area / sc_area
Lesion count = 1
Midsagittal slice of the spinal cord: 10

Measures on lesion #1...
  (S-I) length: 11926.41 mm
  Max. equivalent diameter: 0.41 mm
  Maximum axial damage ratio: inf
  (S-I) length in the midsagittal slice: 8719.91 mm
  (A-P) width in the midsagittal slice: 0.01 mm
  Volume: 727.35 mm^3

Orient output image to initial orientation...
^[[33mFile sub-zh04_ses-02_acq-sag_T2w_lesion_seg_label.nii.gz already exists. Will overwrite it.^[[0m


Averaged measures across all lesions...
  volume [mm3] = 727.35 +/- 0.0
  length [mm] = 11926.41 +/- 0.0
  max_equivalent_diameter [mm] = 0.41 +/- 0.0
  max_axial_damage_ratio [] = inf +/- nan

Minimum tissue bridges across all lesions for the midsagittal slice (sagittal slice 10)...
WARNING: Lesion #1 does not exist in the midsagittal slice
  Minimum dorsal bridge width [mm]: nan
  Minimum ventral bridge width [mm]: nan

Total volume = 727.35 mm^3
Lesion count = 1

Save results files...

... measures saved in the files:

  - ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_label.nii.gz
cp /tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau/sub-zh04_ses-02_acq-sag_T2w_lesion_seg_label.nii.gz ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_label.nii.gz

  - ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.xls
cp /tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau/sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.xls ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.xls

  - ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.pkl
cp /tmp/sct_2024-09-20_12-02-42_analyze-lesion_fwwtdhau/sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.pkl ./sub-zh04_ses-02_acq-sag_T2w_lesion_seg_analysis.pkl

*** Generating Quality Control (QC) html report ***
Traceback (most recent call last):
  File "/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/spinalcordtoolbox/scripts/sct_analyze_lesion.py", line 1085, in <module>
    main(sys.argv[1:])
  File "/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/spinalcordtoolbox/scripts/sct_analyze_lesion.py", line 1048, in main
    sct_analyze_lesion(
  File "/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/spinalcordtoolbox/reports/qc2.py", line 580, in sct_analyze_lesion
    axes[idx_row, idx_col].set_xlim(np.min(np.where(im_label_data_cur)[1]) - 20,
  File "/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/python/envs/venv_sct/lib/python3.9/site-packages/numpy/core/fromnumeric.py", line 2953, in min
    return _wrapreduction(a, np.minimum, 'min', axis, None, out,
  File "/home/GRAMES.POLYMTL.CA/p118175/code/spinalcordtoolbox/python/envs/venv_sct/lib/python3.9/site-packages/numpy/core/fromnumeric.py", line 88, in _wrapreduction
    return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
ValueError: zero-size array to reduction operation minimum which has no identity

The issue comes from L542 in reports/qc2.py, which restricts the lesion mask to the spinal cord mask. This effectively zeros the lesion as it is not covered by the cord seg.

… metal artifacts See comments in the following PR for context: #96

…bjects

valosekj · 2024-10-02T14:45:38Z

As the C-A model does not perform well on lumbar images with metal artifacts, I dropped it in 3a88281. The further analysis will continue with the SCIsegV2 model only.

…ecause if SCIseg predictions no lesion, it finished with a QC error and GT masks are not analyzed at all

valosekj added 4 commits September 20, 2024 11:20

Add script to segment the spinal cord and lesions from T2w images and…

426b718

… compute the midsagittal lesion length and width.

Change logic how to save XLS files produced by sct_analyze_lesion. If…

02e9321

… we use '-ofolder' for sct_analyze_lesion, then QC report fails.

Remove forgotten '-ofolder ${PATH_RESULTS}/contrast-agnostic' from th…

7b87b5e

…e previous commit

Remove 'CUDA_VISIBLE_DEVICES=1 SCT_USE_GPU=1' for the contrast-agnost…

6a06f86

…ic model for now Context: spinalcordtoolbox/spinalcordtoolbox#4633

Use '-largest 1' to keep only the largest connected component. Context:

8d9b7b9

https://github.com/ivadomed/model_seg_sci/pull/96\#issuecomment-2364121461

valosekj added 2 commits September 20, 2024 13:36

Add README for the midsagittal lesion length and width analysis

1f647a6

Generate sagittal lesion QC report

813caae

naga-karthik mentioned this pull request Sep 23, 2024

Contrast-agnostic model on a lumbar SCI patient outputs incorrect segmentation sct-pipeline/contrast-agnostic-softseg-spinalcord#118

Open

valosekj added 2 commits October 2, 2024 10:41

Remove the CA model because it does not work well on lumbar data with…

3a88281

… metal artifacts See comments in the following PR for context: #96

Add 02_read_xls_files.py script to aggregate lesion metrics across su…

b02b11a

…bjects

valosekj added 7 commits October 2, 2024 15:33

Include measurements also for GT masks

80ec10d

Add QC for manual GT

23a1adc

Run sct_analyze_lesion on GT masks before running SCIseg inference. B…

02c33bb

…ecause if SCIseg predictions no lesion, it finished with a QC error and GT masks are not analyzed at all

Add logging to better track number of processed subjects

79fd813

Update docstring

d296080

Take into account multiple lesions

def3d1f

Add script to generate figures

0a7aab6

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Analyze midsagittal lesion length and width #96

Analyze midsagittal lesion length and width #96

valosekj commented Sep 20, 2024 •

edited

Loading

valosekj commented Sep 20, 2024 •

edited by jcohenadad

Loading

valosekj commented Sep 20, 2024 •

edited by jcohenadad

Loading

valosekj commented Oct 2, 2024

Analyze midsagittal lesion length and width #96

Are you sure you want to change the base?

Analyze midsagittal lesion length and width #96

Conversation

valosekj commented Sep 20, 2024 • edited Loading

valosekj commented Sep 20, 2024 • edited by jcohenadad Loading

valosekj commented Sep 20, 2024 • edited by jcohenadad Loading

valosekj commented Oct 2, 2024

valosekj commented Sep 20, 2024 •

edited

Loading

valosekj commented Sep 20, 2024 •

edited by jcohenadad

Loading

valosekj commented Sep 20, 2024 •

edited by jcohenadad

Loading