ICA-reports goals and roadmap

[eurunuela]

Main goal

The goal of ICA-reports is to provide a JavaScript app to visualize and interact with ICA components resulting from different ICA packages (e.g., tedana and aroma). The main goal of ICA-reports is to provide manual classification functionality though.

Making a manual classification app for various ICA algorithms will require a "new" standard for ICA data structure that tedana and aroma (and others) will have to adapt. Obviously, we can build this standard based on the current structure in tedana.

Ideally, ICA-reports should work as a web service, with no need to install the package. React should also allow for the web service to be installed on a machine as a Chrome app.

Roadmap

The following milestones define the roadmap for ICA-reports (feel free to add more):

1. Read ICA components data from a user-provided local .json or .tsv file (we should come up with the standard data structure).
2. Make scatter and pie plots to show rho, kappa, and variance of components.
3. Plot time and frequency signatures of ICA components.
4. Add an onboarding screen.
5. Add support manual classification (new selection + extract as compatible file).
6. Set up a server to run ICA-reports as a web service.

[tsalo]

It would also be cool if we could correlate component time series with arbitrary confounds and generate histograms. Perhaps with support for BIDS physio files and fMRIPrep confounds files?

[eurunuela]

With a number of users having issues with the Bokeh-based reports lately, I believe we should try to release an alpha version of the Bokeh-based reports asap.

I believe the first three milestones are pretty much done. We will have to adapt the way we read the data and make sure the size of the plots is nicer (also include a zooming option). I was thinking we could ask for the JSON file, read it, and assume the images are on the same folder, meaning the tedana outputs weren't moved from the folder. Maps and carpet plots are currently read based on the local path, which is the same path the JSON file is saved in.

@tsalo Re histograms: what did you have in mind?

[tsalo]

I'm thinking of some way to look for outliers on metrics and component-confound correlations, like the group reports for MRIQC. They wouldn't have to be histograms. They could be the box-plot + strip-plot combo that MRIQC uses, for example.

[eurunuela]

As an update: I am very close to replicating the tedana reports in master. What's left before we can consider this an alpha version is:

• Calculating the ranked kappa, rho, and variance values.
• Making the design nicer.

This alpha version could fully replace the tedana reports and would include the first 4 points in the issue description. The beta version would aim at the manual classification of components and extracting the new classification table.