v1.10.x

v1.10.1 (2026-04-20)

New features & enhancements

• Improved performance by caching BIDSPath matches for Maxwell Filter calibration and cross-talk files (#1215 by @larsoner).
• Improved performance and logging when skipping cached steps (#1216 by @larsoner).

Bug fixes

• Fixed a bug where the group average step would errantly write its sensor averaged Evoked plots to the last subject's report. To fix this after-the-fact, delete the derivatives/mne-bids-pipeline/sub-average/ and derivatives/mne-bids-pipeline/sub-<LAST> folders and rerun the pipeline. (#1220 by @larsoner)

v1.10.0 (2026-04-06)

New features & enhancements

• Support for using mne-icalabel to automatically label ICA components. This requires the mne-icalabel package to be installed. (#1018 and #812 by @jschepers, @behinger, @SNaber, @hoechenberger, and @larsoner)
• It is now possible to use separate MRIs for each session within a subject, as in longitudinal studies. This is achieved by creating separate "subject" folders for each subject-session combination, with the naming convention sub-XXX_ses-YYY, in the freesurfer SUBJECTS_DIR. (#987 by @drammock)
• New config option allow_missing_sessions allows to continue when not all sessions are present for all subjects. (#1000 by @drammock)
• New config option mf_extra_kws passes additional keyword arguments to mne.preprocessing.maxwell_filter. (#1038 by @drammock)
• New value "twa" for config option mf_destination, to use the time-weighted average head position across runs as the destination position. (#1043 and #1055 by @drammock)
• New config options mf_cal_missing and mf_ctc_missing for handling missing calibration and cross-talk files (#1057 by @harrisonritz)
• New config options find_bad_channels_extra_kws, notch_extra_kws, and bandpass_extra_kws to pass additional keyword arguments to mne.preprocessing.find_bad_channels_maxwell, mne.filter.notch_filter, and mne.filter.filter_data respectively (#1061 by @harrisonritz)
• Bad channel reports now say which channels were marked as bad in the original data and which were detected automatically by the pipeline. (#1138 by @larsoner)
• Config option ssp_ecg_channel now allows dict values, for setting a different channel name for each subject/session (#1062 by @drammock)
• New config option epochs_custom_metadata allows for custom metadata when creating epochs. (#1088 by @harrisonritz)
• New config option [process_raw_clean] to allow not fully cleaning raw data to save computing resources (#1136 by @larsoner)
• Automatically avoid using joblib.Parallel when only one iteration is needed (and dask is not being used), allowing multithreaded linear algebra operations to be used (#1136 by @larsoner)
• Add support for processing multiple tasks simultaneously using a list like task = ["task1", "task2"] in the config file (#1141 by @larsoner)
• Add support for disabling time-by-time decoding while still performing full-epochs decoding using decoding_time (#1141 by @larsoner)
• Add support for setting decimation for time-by-time decoding using decoding_time_decim (#1141 by @larsoner)
• Add support for subject and session-specific EOG channels in config and change ssp_ecg_channel to accept 'default' key when dict (#1151 by @skjerns)
• Add support for computing data rank using mne.compute_rank via cov_rank (#1159 by @larsoner)
• Add support for setting the number of smoothing steps when averaging via smoothing_steps (#1195 by @larsoner)

Behavior changes

• The pipeline will now raise an error if a loaded SourceSpaces object has a ._subject attribute different from what the pipeline expects / would have used if creating the SourceSpaces anew. (#1056 by @drammock)
• Maxwell filtering steps now use the union of bad channels across all tasks and runs to help ensure more consistent rank and behavior (#1158 by @larsoner)
• The path passed to noise_cov callables is no longer the evoked dataset but instead epochs. Note that the epochs have processing="clean" in the BIDSPath. (#1159 by @larsoner)
• Values within dicts passed to contrasts may now be any sequence (not just lists). (#1197 by @drammock)

Bug fixes

• Empty room matching is now done for all sessions (previously only for the first session) for each subject. (#976 by @drammock)
• noise_cov_method is now properly used for noise covariance estimation from raw data (#1010 by @larsoner)

• When running the pipeline with mf_filter_chpi enabled (#977 by @drammock and @larsoner):

  1. Emptyroom files that lack cHPI channels will now be processed (for line noise only) instead of raising an error.
  2. cHPI filtering is now performed before movement compensation.

• Fix bug where the config.proc parameter was not used properly during forward model creation (#1014 by @larsoner)

• Fix bug where emptyroom recordings containing EEG channels would crash the pipeline during maxwell filtering (#1040 by @drammock)
• Fix bug where only having mag sensors would crash compute_rank during maxwell filtering or epoching (#1061 and #1069 by @harrisonritz)
• Improvements to template config file generation (#1074 by @drammock)
• Fix bug where mf_int_order wasn't passed to maxwell_filter. Added config option for mf_ext_order. (#1092 by @harrisonritz)
• Sanitize report tags that contain " or ', e.g., for certain metadata contrasts (#1097 by @harrisonritz)
• sub-average is no longer run when --subject is passed to the pipeline command line interface, allowing easier and more correct single-subject iteration (#1133 by @larsoner)
• Fix bug where time-weighted head position was not cached properly (#1135 by @larsoner)
• Fix bug where some config options were unnecessarily passed to steps, possibly leading to unnecessary cache misses (#1152 by @larsoner)
• Fix bug where the config.data_type channels were not properly picked during ICA fitting, leading to potential errors during ICA application (#1156 by @larsoner)
• Fix bug where a nonexistent evoked BIDSPath was passed to user-defined noise_cov callables, a path to epochs is now passed instead (#1159 by @larsoner)
• Fix bug where ICA pre-filtering was only applied to run-1, now filters all runs (#1164 by @Okimoka)
• Fix bug where runs were not handled properly when a single task was specified but multiple tasks existed in a dataset (#1141 by @larsoner)

Documentation

• Choose the theme (dark of light) automatically based on the user's operating system setting (#979 by @hoechenberger)
• Bundle all previously-external JavaScript to better preserve users' privacy (#982 by @hoechenberger)
• Document the need for offscreen rendering support when running on headless servers (#997 by @drammock)
• Add more info on test suite installation and running (#1154 by @skjerns)

Code health and infrastructure

• Switch from using relative to using absolute imports (#969 by @hoechenberger)
• Enable strict type checking via mypy (#995, #1013, #1016 by @larsoner)
• Improve logging messages in maxwell filtering steps. (#893 by @drammock)
• Validate extra config params passed in during testing. (#1044 by @drammock)
• New testing/example dataset "funloc" added. (#1045 by @drammock)
• Bugfixes and better testing of session-specific MRIs. (#1039 and #1067 by @drammock)
• Drop legacy function inst.pick_types in favor of inst.pick (#1073 by @PierreGtch)
• Moved to ty, lefthook, and dependency-groups for development (#1166 by @larsoner)