Statistics-Volume pipeline #54

arnaudmarcoux · 2019-11-29T16:54:37Z

This PR adds 2 new pipelines : statistics-volume and statistics-volume-correction

statistics-volume performs a 2 sample t-test using spm, with covariates indicated in the provided tsv
statistics-volume-correction

pep8speaks · 2020-02-11T18:15:20Z

Hello @arnaudmarcoux! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

In the file clinica/pipelines/statistics_volume/statistics_volume_cli.py:

Line 42:9: E303 too many blank lines (2)

In the file test/instantiation/test_instantiate_all_pipelines.py:

Line 405:21: W292 no newline at end of file

In the file test/nonregression/test_run_pipelines.py:

Line 755:5: E303 too many blank lines (2)
Line 783:1: W293 blank line contains whitespace

Comment last updated at 2020-02-28 15:44:05 UTC

clinica/pipelines/statistics_surface/statistics_surface_cli.py

clinica/pipelines/statistics_volume/info.json

clinica/pipelines/statistics_volume/statistics_volume_cli.py

alexandreroutier · 2020-02-20T11:04:30Z

clinica/pipelines/statistics_volume/statistics_volume_cli.py

+        optional.add_argument("-tup", "--threshold_uncorrected_pvalue",
+                              type=float, default=0.001,
+                              help='Threshold to display the uncorrected p-value '
+                                   + '(--threshold_uncorrected_pvalue 0.001).')
+
+        optional.add_argument("-tcp", "--threshold_corrected_pvalue",
+                              type=float, default=0.05,
+                              help='Threshold to display the corrected p-value '
+                                   '(default: --threshold_corrected_pvalue 0.05)')


I think these flags are either not used or have wrong help message. Is this a copy/paste from statistics-surface?

I do not really know what else to say about them (and yes it is a copy paste :)

Ok. Simply remove these lines, there is no equivalent in your code.

clinica/pipelines/statistics_volume/statistics_volume_cli.py

clinica/pipelines/statistics_volume/statistics_volume_pipeline.py

clinica/pipelines/statistics_volume/template_model_contrast.m

clinica/pipelines/statistics_volume/template_model_creation.m

environment.yml

requirements.txt

clinica/pipelines/statistics_volume/statistics_volume_utils.py

clinica/pipelines/statistics_volume/statistics_volume_pipeline.py

alexandreroutier · 2020-02-23T17:51:50Z

@mdiazmel : In this PR, the clinica/resources/mni_icbm152_t1_tal_nlin_sym_09a.nii.gz will be added. Is this the same template you used in T1Linear pipeline in PR #62?

@arnaudmarcoux: There are several flavours of the ICBM template. Could you tell me why you chose this version compared to another variant? If there is no particular reason, I would be in favor of using the one in T1Linear pipeline.

clinica/pipelines/statistics_volume_correction/statistics_volume_correction_pipeline.py

clinica/pipelines/statistics_volume/statistics_volume_utils.py

clinica/pipelines/statistics_volume_correction/statistics_volume_correction_utils.py

alexandreroutier · 2020-02-23T18:02:38Z

clinica/resources/templates/pipeline_template/cli.py.j2

@@ -68,7 +66,7 @@ class {{ pipeline.class_name }}CLI(ce.CmdParser):
        # BIDS and/or CAPS directory as inputs. If the BIDS directory is not needed
        # for your pipeline, simply remove:
        # bids_directory=self.absolute_path(args.bids_directory),
-        pipeline = T1FreeSurfer(
+        pipeline = {{ pipeline.class_name }}(


This is currently fixed in PR #63 .

clinica/pipelines/statistics_volume/statistics_volume_pipeline.py

alexandreroutier · 2020-02-23T18:20:45Z

Hi @arnaudmarcoux,

I finished a first reading of your code. We are quite close to the final version :)

I noticed that instanciation tests are missing. Moreover, I would advice you to move the content of statistics_volume_correction folder to statistics_volume folder. I thinks this make sense since explanations on the statistics-volume / statistics-volume-correction pipelines will be on the same Wiki page.

Btw, it seems that outputs are not temporarly saved in working directory. Coud you tell me if this is the case or not?

I will do the feedback regarding the Wiki tomorrow.

Alex

alexandreroutier · 2020-02-25T09:56:57Z

Currently, the output console looks like:

No info on how much size the pipeline takes. Running anyway...
USING CLASSICAL SPM
USING CLASSICAL SPM
USING CLASSICAL SPM
USING CLASSICAL SPM
USING CLASSICAL SPM
[11:59:36] The statistics-volume pipeline has completed. You can now delete the working directory (/home/alexandre.routier/Temp/StatisticsVolume/in/WD/statistics-volume).

The first line will disappear when info.json will be updated. You should remove or comment the cprint instructions regarding the use of SPM / SPM-Standalone.

In build_input_node() method, you should add something like this:

if len(self.subjects):
    print_images_to_process(self.subjects, self.sessions)
    cprint('The pipeline will last a few minutes. Images generated by SPM will popup during the pipeline.')
    print_begin_image('group-' + self.parameters['group_id'])

(Adapted from t1-create-dartel pipeline)

arnaudmarcoux · 2020-02-25T15:05:24Z

The first line will disappear when info.json will be updated.

Indeed, pipeline disk usage is determined by:
( number or subjects in the analysis) X space_caps and ( number or subjects in the analysis) X space_wd. In this case, disk usage forthis pipeline is always the same and is not dependant on the number of subjects. We cannot use the current system.

You should remove or comment the cprint instructions regarding the use of SPM / SPM-Standalone.

Done

In build_input_node() method, you should add something like this:

if len(self.subjects):
    print_images_to_process(self.subjects, self.sessions)
    cprint('The pipeline will last a few minutes. Images generated by SPM will popup during the pipeline.')
    print_begin_image('group-' + self.parameters['group_id'])

Done

Btw, it seems that outputs are not temporarly saved in working directory. Coud you tell me if this is the case or not?

Everything is working on working dirs, except last node of statistics-volume-correction. I did not have a simple method to save the outputs, so I did not use a datasink. That's why there is no output node.

alexandreroutier · 2020-02-26T11:23:09Z

test/nonregression/test_run_pipelines.py

+    )
+    pipeline.run(plugin='MultiProc', plugin_args={'n_procs': 4}, bypass_check=True)
+    compare_folders(join(root, 'out'), join(root, 'ref'), 'caps')
+


Missing:

# Remove data in out folder clean_folder(join(root, 'out', 'caps'), recreate=False)

?

alexandreroutier · 2020-02-26T11:23:20Z

test/nonregression/test_run_pipelines.py

+    assert np.allclose(nib.load(output_t_stat).get_data(),
+                       nib.load(ref_t_stat).get_data())
+


Missing:

# Remove data in out folder clean_folder(join(root, 'out', 'caps'), recreate=False)

?

alexandreroutier · 2020-02-28T08:25:39Z

When running statistics-volume pipeline, I have the following files;

└── group-GroupTest
    └── statistics_volume
        └── group_comparison_measure-fdg
            ├── group-GroupTest_report-1.png
            ├── group-GroupTest_report-2.png
            ├── group-GroupTest_report-3.png
            ├── group-GroupTest_report-4.png
            ├── group-GroupTest_report-5.png

Why do we have 5 *report.png files ? We should only have 2: group-GroupTest_AD-lt-CN_measure-fdg_fwhm-8_report.png and group-GroupTest_CN-lt-AD_measure-fdg_fwhm-8_report.png containing results of SPM at cluster_threshold of 0.001.

Besides, I made some mistakes when specifying the CAPS:

└── group-GroupTest
    └── statistics_volume
        └── group_comparison_measure-fdg
            ├── group-GroupTest_contrast-group_mask.nii
            ├── group-GroupTest_contrast-group_RPV.nii

_contrast-group should not exist thus giving the final files group-GroupTest_mask.nii and group-GroupTest_RPV.nii.

Finally:

└── group-GroupTest
    └── statistics_volume
        └── group_comparison_measure-fdg
            ├── group-GroupTest_contrast-group_measure-AD_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_contrast-group_measure-age_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_contrast-group_measure-CN_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_contrast-group_measure-sex_fwhm-8_regressionCoefficient.nii

It should be _measure-fdg in all cases. When I put contrast-<contrast> at the beginning, I was thinking of contrast-age or contrast-sex for instance. In hindsight, I would suggest instead covariate-age giving the final files:

└── group-GroupTest
    └── statistics_volume
        └── group_comparison_measure-fdg
            ├── group-GroupTest_covariate-AD_measure-fdg_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_covariate-age_measure-fdg_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_covariate-CN_measure-fdg_fwhm-8_regressionCoefficient.nii
            ├── group-GroupTest_covariate-sex_measure-fdg_fwhm-8_regressionCoefficient.nii

Open question:

In t1-volume / pet-volume pipelines, smoothing values are reported like _fwhm-8mm whereas pet-surface / statistics-surface report smoothing like _fwhm-10. In this pipeline, _fwhm-8 convention is used. Do we keep it?

alexandreroutier · 2020-02-28T08:27:21Z

Final request: can you move the content of statistics_volume_correction into statistics_volume ?

alexandreroutier · 2020-02-28T08:39:34Z

The first line will disappear when info.json will be updated.

Indeed, pipeline disk usage is determined by:
( number or subjects in the analysis) X space_caps and ( number or subjects in the analysis) X space_wd. In this case, disk usage forthis pipeline is always the same and is not dependant on the number of subjects. We cannot use the current system.

Space estimation was given in t1-volume-create-dartel even though it is a groupe pipeline. Do you you remember how the values were set? In order to apply the same rationale for your pipeline.

arnaudmarcoux · 2020-02-28T14:34:20Z

Space estimation was given in t1-volume-create-dartel even though it is a groupe pipeline. Do you you remember how the values were set? In order to apply the same rationale for your pipeline

I don't remember how this was set...

Why do we have 5 *report.png files ?

2 different contrasts group 1 > group 2 and group 2 > group 1
And for each (SPM settings) : 1) p=0.05 FWE (that is traditional use of SPM) and 2) p=0.001 uncorrected

Which makes 4 reports (5 because one of them is more than 1-page long. Moreover It is not possible to link each report to its task.

_contrast-group should not exist thus giving the final files group-GroupTest_mask.nii and group-GroupTest_RPV.nii.

Done

In t1-volume / pet-volume pipelines, smoothing values are reported like _fwhm-8mm whereas pet-surface / statistics-surface report smoothing like _fwhm-10. In this pipeline, _fwhm-8 convention is used. Do we keep it?

I dont really have an opinion

arnaudmarcoux · 2020-02-28T14:50:44Z

Also, I did this image a while ago, if you want to use it in the wiki to indicate where to look for statistics-volume-correction. All the items marked are necessary to determine what to indicate in the command line interface.

alexandreroutier · 2020-02-28T14:58:13Z

I don't remember how this was set...

Ok. We'll put that aside.

2 different contrasts group 1 > group 2 and group 2 > group 1
And for each (SPM settings) : 1) p=0.05 FWE (that is traditional use of SPM) and 2) p=0.001 uncorrected

Which makes 4 reports (5 because one of them is more than 1-page long. Moreover It is not possible to link each report to its task.

Ok. Since FWEp is performed in statistics-volume-correction, it would be simpler to remove reports regarding FWEp. What do you think?

alexandreroutier · 2020-02-28T15:02:46Z

Also, I did this image a while ago, if you want to use it in the wiki to indicate where to look for statistics-volume-correction. All the items marked are necessary to determine what to indicate in the command line interface.

Thanks! I put a report in the Wiki but it didn't have annotations.

alexandreroutier · 2020-03-03T17:25:29Z

To be followed in PR #67.

alexandreroutier added enhancement New feature or request pipeline labels Dec 6, 2019

alexandreroutier added this to the 0.4.0 milestone Jan 22, 2020

alexandreroutier self-requested a review February 20, 2020 10:22