Tedana February 2025 Update
tedana-newsletter
Multi-echo fMRI users meeting!
The tedana group is planning to start hosting periodic multi-echo fMRI users meetings. The goal is to gather online several times per year to discuss issues relating to multi-echo fMRI. This can include practical ask-me-anything sessions, discussions of best practices or challenges, presentations of works-in-progress, and presentations of published work. While tedana developers are initiating this effort, we welcome topics that aren't directly about tedana and we welcome others who might want to help organize and plan. Please take this survey to help us gauge interest and schedule our first meeting: https://forms.gle/C2NHNNxmGehXw9oS7
Invite people to this newsletter
The multi-echo fMRI community and the tedana users community is growing rapidly and is substantially larger than the number of people subscribed to this group. Please encourage other tedana and multi-echo fMRI users to this low frequency newsletter at: https://groups.google.com/g/tedana-newsletter
tedana version 24.0.2
We have added multiple new features since our last newsletter for version 24.0.0
A long requested feature has been to include external time series into the tedana’s decision process. 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 user 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 #1064.
Another long-standing challenge has been variable ICA results on different systems and some failures in the number of estimated components. With the help of new contributors Bahman Tahayori and Robert Smith, we have added an option to use RobustICA. This runs FastICA (which we have always used) multiple times and outputs more stable components. As part of this process, if the PCA step estimates X components, RobustICA often finds fewer stable components and will output fewer than X ICA components. As we’ve previously noted, the automated estimates for the number of PCA components sometimes fails. With the addition of RobustICA, if the automated estimate fails, one can input a plausible, but slightly high fixed number of PCA components (i.e. “--tedpca 70” for 70 components), and RobustICA will identify a lower number of stable ICA components. This leads 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. #1013.
We create an adaptive mask that lists the number of "good echoes" in each voxel. This is used so that a voxel with a single good echo is retained, but steps that include fitting values across echoes are limited to voxels with more good echoes. We noticed a few places where our description of how the adaptive mask was created didn't match what was actually happening. These were bugs that needed to get fixed, but our underlying thresholds are arbitrary. If these fixes are adversely affecting the retained voxels, please speak up and we can examine tweaking thresholds or adding other options#1060 & #1061. A new adaptive masking is described in #1057
We have also added several long-requested visualizations to our html report. Descriptions of the new visualizations are in our documentation. The additions include:
A visualization of the adaptive mask to show which voxels are retained in the optimally combined image and which are used T2* and S0 fits as part of the ICA denoising process. #1073
Mean T2* and S0* fits are calculated and used as weights for the optimal combination of echoes. The root mean square error (RMSE) for this fits are now saved and presented in our report #1044
The full release notes for versions 24.0.1 and 24.0.2 are at: https://github.com/ME-ICA/tedana/releases
Active work
Improving methods to robustly estimate the total number of components
Now that they are added, identifying new ways to use external regressors in the component selection process
Improving visualizations and quality checks of results, particularly with the newly added RobustICA option
Getting help with tedana or multi-echo fMRI
Questions about multi-echo fMRI and tedana usage or development can be posted at https://neurostars.org with ‘tedana’ or ‘multi-echo’ tags or as an issue or discussion at https://github.com/ME-ICA/tedana or https://mattermost.brainhack.org/brainhack/channels/tedana. We actively monitor all three message boards and try to efficiently respond.
Contributors
We are excited to welcome new contributors: @BahmanTahayori, @Lestropie, and @mvdoc. We continue to thank all of our contributors for their continued input and help with tedana. We always look forward to seeing more new faces! If you’re not sure where to start, please feel free to open an issue on github, ask a question on neurostars with the ‘tedana’ or ‘multi-echo’ tag, join our next monthly developer’s call, or join the tedana mattermost channel.