sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline #2020

charleygros · 2018-09-10T03:54:53Z

Rational:
In some cases, the centerline detection (svm- or cnn-based) fails, and it would be nice to be able to bypass this step by inputing our own centerline (e.g., obtained semi- or fully-manually).

New features:
A new option in the flag "centerline", where a file name or "viewer" can be input.
Example of usage with the viewer:

sct_deepseg_lesion -i t2.nii.gz -c t2 -centerline viewer

Example of usage with the manual centerline:

sct_deepseg_sc -i t2.nii.gz -c t2 -centerline t2_centerline_manual.nii.gz

Approach:

For the viewer: it uses the new version of the viewer and then use NURBS to interpolate the viewer_labels to a continuous centerline, with the following parameters: algo_fitting='nurbs', nurbs_pts_number=10000.
For the manual centerline: we applied the same preprocessing steps as to the input image.

Related issues:

jcohenadad · 2018-09-10T14:56:50Z

scripts/sct_deepseg_lesion.py

@@ -51,14 +51,14 @@ def get_parser():
 mandatory=True,
 example=['t2', 't2_ax', 't2s'])
 parser.add_option(name="-centerline",
- type_value="multiple_choice",


you want to keep "multiple_choice", otherwise the values "cnn" and "svm" will be considered as a type:nifti_input wheres they should be considered as type:string by the Parser.

related to #2020 (comment)

My answer here: #2020 (comment)

jcohenadad · 2018-09-10T15:21:14Z

scripts/sct_deepseg_lesion.py

- type_value="multiple_choice",
- description="choice of spinal cord centerline algorithm.",
+ type_value="image_nifti",
+ description="Method used for extracting the centerline.\nsvm: automatic centerline detection, based on Support Vector Machine algorithm.\ncnn: automatic centerline detection, based on Convolutional Neural Network.\nviewer: semi-automatic centerline generation, based on manual selection of a few points using an interactive viewer, then approximation with NURBS.\nprovide the filename of a manual centerline (e.g. t2_centerline_manual.nii.gz).\n",


This part of the doc is a bit confusing wrt. input file name (it is not clear what one should write after -centerline for inputing a file name. I would rewrite as:

type_value="multiple_choice", description="Method used for extracting the centerline. "\nsvm: automatic centerline detection, based on Support Vector Machine algorithm." "\ncnn: automatic centerline detection, based on Convolutional Neural Network." "\nviewer: semi-automatic centerline generation, based on manual selection of a few points using an interactive viewer, then approximation with NURBS." "\nFILENAME: filename of an existing centerline (e.g. t2_centerline_manual.nii.gz).\n",

Now, the problem is that my suggestion above will NOT work, because multiple_choice will look for the string FILENAME instead of an actual file name :-(

I see two alternatives:

replace FILENAME by manual and have another flag dedicated to inputing a centerline file, e.g.: sct_deepseg_sc -i t2.nii.gz -c t2 -centerline manual -filecenterline t2_centerline_manual.nii.gz. Cons: heavy.

modify the Parser to have the possibility to not check the input string when using multiple_choice. Cons: defeat the purpose of a parsing checker...

@zougloub any idea?

@jcohenadad: I initially tried with multiple_choice and I faced the same issue as you just mention above.
As you said, I used type:image_nifti to allow the user to input a manul centerline. While the other options (i.e. svm, cnn, viewer) are not considered as Image, from my understanding, thanks to the parameter list_no_image=['svm', 'cnn', 'viewer'].
--> This approach is inspired from PropSeg:
https:/neuropoly/spinalcordtoolbox/blob/master/scripts/sct_propseg.py#L242

Part of the doc corrected as suggested by @jcohenadad, in commit #f34bf6b.

the problem is that the way this is currently done in PropSeg is not good either (for the same reasons as here). It would be nice to find a proper solution and then deploy it to the rest of SCT. Waiting for @zougloub's feedback there.

@jcohenadad : @zougloub approved the PR, should I merge the branch or wait for further discussions about the point you raised?

i suggest the following:

"svm: automatic centerline detection, based on Support Vector Machine algorithm." "cnn: automatic centerline detection, based on Convolutional Neural Network." "viewer: semi-automatic centerline generation, based on manual selection of a few points using an interactive viewer, then approximation with NURBS." "manual: filename of an existing centerline (e.g. t2_centerline_manual.nii.gz).\n",

and another flag:

-file-centerline; type:type:image_nifti; description: Input centerline file (to use with flag -centerline manual)

@jcohenadad : I have made the modifications we discussed yesterday.
Ready to be reviewed.

jcohenadad · 2018-09-10T15:22:46Z

scripts/sct_deepseg_sc.py

+ # average centerline coordinates over slices of the image
+ x_centerline_fit_rescorr, y_centerline_fit_rescorr, z_centerline_rescorr, x_centerline_deriv_rescorr, y_centerline_deriv_rescorr, z_centerline_deriv_rescorr = centerline.average_coordinates_over_slices(image_labels)
+
+ # compute z_centerline in image coordinates for usage in vertebrae mapping


not sure i see the link with vertebral mapping here?

Corrected in commit #86fa0db

zougloub

Some questions

zougloub · 2018-09-12T02:42:37Z

scripts/sct_deepseg_lesion.py

@@ -190,6 +190,9 @@ def deep_segmentation_MSlesion(fname_image, contrast_type, output_folder, ctr_al
 tmp_folder = sct.TempFolder()
 tmp_folder_path = tmp_folder.get_path()
 fname_image_tmp = tmp_folder.copy_from(fname_image)
+ if os.path.isfile(ctr_algo): # if the ctr_algo is a manual centerline file
+ ctr_algo = os.path.basename(ctr_algo)


why take the base name?

For the same reason as:
https:/neuropoly/spinalcordtoolbox/blob/master/scripts/sct_deepseg_lesion.py#L189
--> If we pass the if condition, then the ctr_algo is the filename of a nifti mask.

Do you recommend to change the name of the variable (initially this variable reffered to algorithm name)? to change the if condition (ie check if it is a nifti image)?

I just want to know why you are trimming the path to its basename, why not keep a full path or absolute path. What if the file is in another folder?
Note: I just took like 1 minute to do a very quick review your code, I didn't look at the implications, just asking.

hehe! very good point thank you.
I swap the two lines (see commit ccd068a) : now I copy the file in the tmp folder then I extract the basename --> which will be usefull in the remaining of the code when processing in the tmp_folder.

zougloub · 2018-09-12T02:43:37Z

scripts/sct_deepseg_sc.py

@@ -136,7 +136,7 @@ def crop_image_around_centerline(filename_in, filename_ctr, filename_out, crop_s
 x_lst, y_lst = [], []
 data_im_new = np.zeros((crop_size, crop_size, im_in.dim[2]))
 for zz in range(im_in.dim[2]):
- if 1 in np.array(data_ctr[:, :, zz]):
+ if np.sum(np.array(data_ctr[:, :, zz])):


could also use np.any for more semantic

… into cg_deepseg_viewer_centerline

…ropoly/spinalcordtoolbox into cg_deepseg_viewer_centerline merge

jcohenadad · 2018-09-28T15:16:56Z

scripts/sct_deepseg_lesion.py

- sys.exit(1)
+ if "-file_centerline" in args:
+ manual_centerline_fname = arguments["-file_centerline"]
+ if ctr_algo != 'manual':


alternatively we can automatically set ctr_algo='manual' (or ignore that field) if user enters a manual centerline. It will make lighter syntax.

Good suggestion, thank you. Done in the last commit.

sct_deepseg_sc/lesion: allow to input manual or semi manual centerline Former-commit-id: b33a7b3

charleygros added 3 commits September 10, 2018 11:26

update parser documentation

0c75d4e

allow to use viewer for generating the centerline

53b0b04

allow to use manual centerline as input

de87919

charleygros self-assigned this Sep 10, 2018

charleygros added sct_deepseg_lesion context: sct_deepseg_sc context: labels Sep 10, 2018

charleygros added this to the v3.2.4 milestone Sep 10, 2018

fix issues of parser for centerline param

4350f0d

charleygros requested review from zougloub and jcohenadad September 10, 2018 03:57

charleygros added 2 commits September 10, 2018 14:01

fix typo

5930be3

improve parser doc

c9093cb

jcohenadad reviewed Sep 10, 2018

View reviewed changes

charleygros added 2 commits September 11, 2018 08:48

improve description of -centerline flag

f34bf6b

remove vertebral mapping, useless in this application

86fa0db

zougloub reviewed Sep 12, 2018

View reviewed changes

charleygros added 2 commits September 12, 2018 13:07

use np.any instead of np.sum for more semantic

d3af3c3

copy in tmp folder before to take the basename

ccd068a

jcohenadad modified the milestones: v3.2.4, v3.2.5 Sep 12, 2018

zougloub approved these changes Sep 15, 2018

View reviewed changes

zougloub and others added 5 commits September 14, 2018 21:45

Merge branch 'master' into cg_deepseg_viewer_centerline

6c7990c

add flag file_centerline in parser

015c9c6

allow user to input centerline file

a863dde

Merge branch 'master' of https:/neuropoly/spinalcordtoolbox…

2de2ef1

… into cg_deepseg_viewer_centerline

Merge branch 'cg_deepseg_viewer_centerline' of https:/neu…

0efb1e8

…ropoly/spinalcordtoolbox into cg_deepseg_viewer_centerline merge

charleygros added the card:TO_REVIEW label Sep 28, 2018

fix error when file_centerline not provided

032fa99

jcohenadad reviewed Sep 28, 2018

View reviewed changes

set ctr_algo=manual if file_centerline provided

22a5eee

jcohenadad approved these changes Sep 30, 2018

View reviewed changes

charleygros merged commit b33a7b3 into master Oct 1, 2018

jcohenadad added feature category: new functionality and removed card:TO_REVIEW labels Oct 16, 2018

jcohenadad changed the title ~~sct_deepseg_sc/lesion: allow to input manual or semi manual centerline~~ sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline Oct 16, 2018

gmotzespina deleted the cg_deepseg_viewer_centerline branch September 13, 2019 08:04

jcohenadad pushed a commit that referenced this pull request Dec 18, 2019

Merge pull request #2020 from neuropoly/cg_deepseg_viewer_centerline

294801d

sct_deepseg_sc/lesion: allow to input manual or semi manual centerline Former-commit-id: b33a7b3

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline #2020

sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline #2020

charleygros commented Sep 10, 2018 •

edited

Loading

jcohenadad Sep 10, 2018 •

edited

Loading

jcohenadad Sep 10, 2018

charleygros Sep 10, 2018

jcohenadad Sep 10, 2018

charleygros Sep 10, 2018 •

edited

Loading

charleygros Sep 10, 2018

jcohenadad Sep 11, 2018

charleygros Sep 16, 2018

jcohenadad Sep 27, 2018

charleygros Sep 28, 2018

jcohenadad Sep 10, 2018

charleygros Sep 10, 2018

zougloub left a comment

zougloub Sep 12, 2018

charleygros Sep 12, 2018

zougloub Sep 12, 2018

charleygros Sep 12, 2018

zougloub Sep 12, 2018

charleygros Sep 12, 2018

jcohenadad Sep 28, 2018

charleygros Sep 30, 2018

sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline #2020

sct_deepseg_sc/lesion: Allow to input manual or semi-manual centerline #2020

Conversation

charleygros commented Sep 10, 2018 • edited Loading

jcohenadad Sep 10, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

charleygros Sep 10, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zougloub left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

charleygros commented Sep 10, 2018 •

edited

Loading

jcohenadad Sep 10, 2018 •

edited

Loading

charleygros Sep 10, 2018 •

edited

Loading