Skip to content

24.0.2

Latest
Compare
Choose a tag to compare
@tsalo tsalo released this 23 Sep 16:50
· 1 commit to main since this release
df58347

Release Notes

Enhancements

  • Generate metrics from external regressors using F stats by @handwerkerd in #1064.
    We have added in a long-requested addition to functionality. It is now possible to provide a TSV file with time series the same length as the fMRI time series, fit those time series to each ICA component, and use that information in the process to decide which components to accept or reject. For example, it is possible to use head motion regressors, cardiac and respiratory, regressors, and region-of-interest based regressors in models. With this functionality, it is now possible to combine the echo-based methods of tedana with other ICA-based denoising methods that depend on fitting to time series. Best practices for how to apply this new functionality are still a work-in-progress, but by adding this functionality, any use can start testing and contributing to this effort without needing to edit code. More information is available in https://tedana.readthedocs.build/en/stable/building_decision_trees.html#external-regressor-configuration

  • Adding robustica option to ICA decomposition to achieve consistent results by @BahmanTahayori in #1013.
    tedana previously used a single iteration of FastICA. This works but it means that the results are sensitive to initial seed selection. We have added in an option to use RobustICA which runs FastICA multiple times and outputs more stable components. As part of this process, if the PCA step defines, X components, robustica often finds fewer stable components and will output fewer than X ICA components. A benefit of this is that our PCA-based component estimation methods sometimes fail. By giving robustica a plausible number of PCA components, it will find a stable number of ICA components leading to a more stable and less arbitrary result. We are still working on improving the stability of the step that initially defines the number of PCA components.

🐛 Bug Fixes

  • Use nearest-neighbors interpolation in plot_component by @tsalo in #1098
  • Filter out non-diagonal affine warning by @tsalo in #1103
  • Refactor gscontrol module by @tsalo in #1086

Documentation Changes

Other Changes

  • Refactor metrics.dependence module by @tsalo in #1088

New Contributors

Full Changelog: 24.0.1...24.0.2