Revert 280 and 281, update customization names and documentation #291
Conversation
Reverts AIM-Harvard#280 / AIM-Harvard#281, as this change introduces some confusion into the what goes into `**kwargs` for initialization of featureextractor (or one of the feature classes). `**kwargs` only contains settings (i.e. one of the three categories of customization). PRs AIM-Harvard#280 and AIM-Harvard#281 included `inputImages` (one of the other categories) to enable customizing the enabled input image types in the slicer extension. However, the functionality to customize this is already present in featureextractor through function calls (such as `enableAllInputImages()`). Therefore, revert these PRs and update the slicer extension to make use of this API.
JoostJM
changed the title
Make distinction between different categories of parameters more explicit, and rename categories to better reflect their definition.
PyRadiomics can be customized via 3 categories of *parameters*:
1. "imageType": specifies filters / image types that can be used to extract features from
2. "featureClass": specifies which feature classes and features should be extracted
3. "setting": specifies all setting to customize how the extraction is done.
In PyRadiomics, "parameter" refers to one or more customization categories and "setting" refers explicitly to type 3 customization.
Furthermore, document that only type 3 parameters can be provided at initialization when not using the parameter file (type 1 and 2 have to be changed using dedicated functions in featureextractor (type 1 and 2) or the individual feature classes (type 2).
Especially for type 1, this includes renaming of various variables, including the name used in the parameter file ("imageType" in stead of "inputImage").
Finally, expand documentation on making the parameter file, including an example directly in the documentation (instead of just referring to the repository).
| "\t original_glcm_DifferenceAverage : 5.58932678922\n", | ||
| "\t original_glcm_Id : 0.276140402104\n", | ||
| "\t original_glcm_ClusterTendency : 103.142793792\n", | ||
| "\t original_glcm_SumVariance : 108.731393255\n", |
There was a problem hiding this comment.
It is not obvious from the title and comments in the PR that it should lead to a change in the feature baselines. Is this expected?
There was a problem hiding this comment.
@fedorov, only the values for GLCM are different in the notebook, other values have just been moved.
As to the difference in GLCM values, that has to do with the bug that was fixed in #261, which could cause the GLCM to be incorrect, depending on your hardware and size of the input.
It took me a while to find out, as I got correct values (equal the 'new' values in this PR) on my regular computer, even when using PyRadiomics version 1.1.1.post80.dev0+g215030b. Turns out that this computer is big enough to add the transposed matrix in one go for brain1 (therefore not causing bug as in #261), but my laptop isn't.
Baseline for GLCM was generated using my regular computer, therefore these were correct, even prior to #261, but the notebooks were generated using my laptop, with different values for GLCM as a result (last update of this notebook was prior to #261).
|
@JoostJM note that SlicerRadiomics uses https://github.com/Radiomics/pyradiomics/tree/slicer branch, you will need to make the corresponding change there too. |
|
Other than the two comments above, looks good to me, great to see improved documentation! |
True, I will merge this branch into the slicer branch (I didn't want to make 2 PRs, I'll reference this PR in the merge commit) |
Works with new slicer branch (70665cc), main changes necessary are due to renamed functions/variables due to PyRadiomics PR [#291](AIM-Harvard/pyradiomics#291).
#291 PR 291 changed the name of the function, but failed to correct the reference to it in the developers section of the documentation.
Reverts #280 and #281, as these PRs introduced a change which confuses the distinction between the 3 types of customization in PyRadiomics.
Furthermore, renames variables and updates documentation to make this distinction more clear.
Throughout PyRadiomics, "parameters" refers to one or more categories of customization in PyRadiomics (i.e. all customization possible). The 3 categories of parameters are:
In the parameter file, all 3 categories can be specified. When initializing the feature extractor without a parameter file, or when using the feature classes directly, only type 3 customization ("setting") can be provided. Changes of type 1 and 2 are applied after intialization, through dedicated function calls (e.g.
enableAllFeaturesandenableAllImageTypesin feature extractor).Includes updates to the documentation to further clarify this, and include a small example for the structure of the parameter file.
cc @Radiomics/developers