slicersalt / slicerdwd

3D Slicer extension for Shape Analysis with DWD (Distance Weighted Discrimination)

Slicer SALT site template

About

Project is built on top of Jekyll - simple, blog-aware, static site generator. Jekyll also happens to be the engine behind GitHub Pages, which means you can use Jekyll to host your website from GitHub’s servers for free. Learn more about Jekyll.

Local development

Check if you have all requirements for local environment or consider using this docker image.

To install all development dependencies install Bundler.

    gem install bundler

and run next command from root folder:

  bundle install

To start Jekyll run:

    jekyll serve -w

Site will be available at http://127.0.0.1:4000/directory/ or http://localhost:4000/directory/ (on Windows)

NOTE: in this mode all changes to html and data files will be automatically regenerated, but after changing _config.yml you have to restart server.

Sass(Compass) support

Note: You need to install Node.js

To watch changes of .sass files and compile it to the .css on a fly change property safe: true to safe: false in _config.yml. Note: It works only on local machine, because GitHub runs Jekyll in --save mode

Learn more about Sass development from documentation.

Resource optimizations (optional)

You can optimize images and minify css and javascript automatically (for now only on Windows). But for Mac OS users available amazing tool - imageoptim. Thanks @raphaelsavina for link. Optimize all images by running this script from /automation/images/ folder:

    all_image_optimization.bat -d -jtran -pout -pquant -optip -gsicle -svgo

To minify CSS and JS run minify_js.bat (for Windows) and minify_js.sh (for Linux and MacOS) from /automation/minifying/ folder:

    minify_js.bat

Learn more about available optimization options from documentation.

Documentation

Quick-start guide is not enough? Checkout full documentation.

Contributors

• Design and web development: Oleh Zasadnyy
• Idea: Vitaliy Zasadnyy

See list of contributors

Maintainers: @tasomaniac and @ozasadnyy.

License

Project is published under the MIT license. Feel free to clone and modify repo as you want, but don't forget to add reference to authors :)

slicerdwd's Issues

Harden installed `dwd` version

Currently dwd is installed with slicer.util.pip_install when the module is initialized; it would probably be better to package it at build-time with CMake.

SlicerDWD/DWD/DWD.py

Lines 15 to 19 in 875cd9a

	try:
	import dwd
	except ModuleNotFoundError:
	slicer.util.pip_install('dwd')
	import dwd

Issues installing `cvxpy` dependency.

Since cvxpy requires development tools be installed on the target machine, it is unsuitable for use in SlicerDWD. It's likely possible to build cvxpy in a superbuild for this module, however after spending some time researching this I've found it will be easier to simply use the existing GenDWD classifier as it does not depend on cvxpy.

See related discussion in https://github.com/slicersalt/dwd/issue/1.

To enable this I've integrated slicersalt/dwd#6, which makes cvxpy an optional dependency (installable via pip install dwd[socp]) and disables the DWD conic solver if it is not installed.

---

Remaining work is to upgrade dwd to 1.0.5 in SlicerDWD, and port to the GenDWD classifier.

Refactor utilities

Currently several functions have been dumped in SlicerDWDLogic that don't really belong.

• Dataset utilities buildCases, splitCases, make_xy
• IO utilities read_vtk, save_vtk, copy_dsa
• Plotting utilities table, scatterPlot, show

SlicerDWDWidget.btnKDEClicked is also very cluttered; much of its behavior should probably be refactored into the same group as the other plotting utilities.

Rename python source to `SlicerDWD.py`

Importing dwd will fail on Windows since the module source is called DWD.py. Rename that so the case-insensitive module names will be different.

Add module documentation/tutorial

• Add tooltips for module fields/buttons
• Describe input formats in documentation/help (csv schema, hyperparameters, etc)
• Include a sample dataset + tutorial for testing

Change "Distance" units to "standard deviation"

Tweaking the distance units to show as "Standard Deviation (&sigma)" would be more intuitive. While technically the distance is a Euclidean distance in mm, it is in a high-dimensional space so shouldn't really be interpreted as such. Regarding it as statistical significance would probably be more meaningful.

Sort samples by DWD distance

Enable sorting samples by distance to the separating hypeplane as in this mock-up:

Since ShapePopulationViewer preserves the order of samples, it should be straightforward to use it to create an initial implementation.

Add README with Screenshots.

Add a README.md with brief description, acknowledgements, and screenshots.

Support train-test-split

Currently all actions (training, analysis) operate on the same dataset. It would be good to implement a way to split training and testing data for more meaningful metrics. Should be able to randomly split a given dataset or manually supply two datasets.

There are already hidden placeholder UI elements, see diff: 875cd9aa

Create Documentation/Tutorial

Create some documentation to allow new users to train a classifier and perform analysis using the module.

Invoke ShapePopulationViewer to display samples

Once a model has been created, allow samples to be opened in ShapePopulationViewer similar to the "preview" modes of ShapeVariationAnalyzer.

The goal is to see something like:

See comments here:

SlicerDWD/DWD/DWD.py

Lines 152 to 175 in 875cd9a

	def meanClicked(self):
	"""Called when the 'Compute Mean" button is clicked."""
	mean = self.logic.meanShape(self.trainCases, factor=50.0)
	model = slicer.mrmlScene.AddNewNodeByClass('vtkMRMLModelNode')
	model.CreateDefaultDisplayNodes()
	model.SetAndObservePolyData(mean.VTKObject)
	model.GetDisplayNode().SetVisibility(True)
	mgr = slicer.app.layoutManager()
	mgr.setLayout(slicer.vtkMRMLLayoutNode.SlicerLayoutOneUp3DView)
	## Ideally this would Launch the CLI ShapePopulationViewer
	## See SVA for how to invoke SPV.
	## https://github.com/NIRALUser/ShapeVariationAnalyzer/blob/master/ShapeVariationAnalyzer/ShapeVariationAnalyzer.py#L742-L764
	## Fails with "Could not find logic for module 'ShapePopulationViewer'"
	## Can manually select the model in SPV though.
	# parameters = {}
	# parameters["vtkFiles"] = mean.VTKObject
	# module = slicer.modules.shapepopulationviewer
	# slicer.cli.run(module, None, parameters, wait_for_completion=True)

And the referenced SVA code.

Ideally the "Distance" scalar would be pre-selected and rendered as vectors, however I don't think that is possible through the SPV CLI.

Add interactive slider to view DWD "mode".

Add an interactive slider as in ShapeVariationAnalyzer and AnnularShapeAnalyzer to move mean shape along the DWD direction.

Enable clicking on plot markers to view samples.

Clicking on a marker in the various plots should load the corresponding sample for viewing.

Use more meaningful names for KDE plot.

The plot legend does not show meaningful names and the axes are poorly labeled.

Show accuracy after training/testing.

Show accuracy, confusion matrix, mean distance, etc. after training or testing data. Related to #3.

There are already hidden placeholder UI elements, see diff: 875cd9aa

Allow computing correlations against other datasets

Should be able to import a CSV of numerical data and plot correlations of the DWD Distance against columns in that dataset.

The goal is to see something like:

If using the Slicer interactive plots widgets: Selecting a group of samples should allow to view in SPV. See #1.