Version 1.14.0.dev175+g357a25258 (development) (2026-11-12)

Dependencies

• Updated minimum for:

  • Optional dependency nibabel >= 5.3

  Changes made automatically by mne[bot]. (#14405)

• Updated minimum for optional dependency h5io >= 0.2.6, which fixes arbitrary code execution when reading crafted HDF5 files with functions such as mne.read_bem_solution() and mne.time_frequency.read_tfrs() (CVE-2026-120499), by Eric Larson. (#14499)

Bugfixes

• Fix bug where plotting covariance matrices created by mne.make_ad_hoc_cov() would raise an IndexError, by new contributor Arnav Kumar and Stefan Appelhoff. (#13582)

• Fix mne.Epochs.export() producing invalid EEGLAB event latencies and dummy events after epochs were dropped, by new contributor Daria Agafonova. (#14067)

• Fix bug where mne.io.RawArray kept a reference to the original data array after operations like resampling or cropping, which wasted memory, by new contributor Aditya Pandey. (#14360)

• Fixed bug where mne.preprocessing.infomax() with extended=True and return_n_iter=True always returned max_iter. It now reports the number of iterations it actually performed, and mne.preprocessing.ICA stores this in the n_iter_ attribute, by new contributor Aslan Satary Dizaji. (#14366)

• Fix merging of overlapping left- and right-eye events in mne.io.read_raw_eyelink() with find_overlaps=True when the events were not already sorted by time, by new contributor Judith Schepers and Scott Huberty. (#14383)

• Fix bug where mne.coreg.Coregistration.set_info() did not clear the omitted head shape points from a previous digitization, by new contributor Badri Vishal Pandey. (#14458)

• Fix bug where mne.preprocessing.compute_average_dev_head_t() averaged head rotation with squared time weights instead of linear ones, by new contributor Ziqi Chen. (#14496)

• Fix bug in mne.time_frequency.psd_array_multitaper() where normalization="full" incorrectly scaled results by 1/sfreq when output="complex", by new contributor Mark Kuang. (#14497)

• Prevent average="mean" and average="median" being used with output="complex" in mne.time_frequency.psd_array_welch(), which incorrectly mixes phases across segments, by new contributor Mark Kuang. (#14498)

• Fix bug where mne.viz.plot_volume_source_estimates() with mode="glass_brain" always showed absolute values, by Lau Møller Andersen. (#13989)

• Fix mne.time_frequency.psd_array_welch() (and Welch-method compute_psd) so that good data spans shorter than n_per_seg no longer raise noverlap must be less than nperseg; such spans are now analyzed with a window shrunk to the span length (with a warning that their spectral resolution is reduced) instead of being lost, by Cedric Conday. (#14003)

• Make docstrings resolve properly in IDEs by fully writing them out in the Python code, by Eric Larson. (#14195)

• Exclude pre-existing bad channels that are not bridged from interpolate_bridged_electrodes() interpolation, by Deepesh Sonar. (#14287)

• Fix reading EEGLAB .set files containing MATLAB objects (e.g., string or datetime) with SciPy 1.18+ when pymatreader is not installed, by Eric Larson. (#14293)

• Fix mne.io.Raw.export() to EDF/BDF altering channel values for data originally read from EDF/BDF files, and add a digital_range parameter to allow preserving the original digital range, by Daniel McCloy. (#14296)

• Fix bug where mne.beamformer.make_lcmv() would create complex rather than real filters when using NumPy 2.5, by Eric Larson. (#14300)

• Fix bug where right-clicking a second component in mne.preprocessing.ICA.plot_sources() raised an IndexError, by Johannes Herforth. (#14303)

• Fix several problems with flat surfaces in mne.viz.Brain: 1. They could rotate incorrectly with the arrow keys or mne.viz.Brain.show_view(), 2. lose their 2D interaction mode when interaction was enabled, 3. orientation/silhouette options either did nothing or caused errors 4. triangle adjacency was calculated incorrectly, by Payam Sadeghi-Shabestari. (#14308)

• Fix bug where mne.filter.notch_filter() and mne.io.Raw.notch_filter() with method="spectrum_fit" failed to detect and remove line noise whose frequency fell between FFT bins, and subtracted a beating artifact from neighboring bins, by Eric Larson. (#14310)

• Fix bug where mne.preprocessing.oversampled_temporal_projection() and mne.preprocessing.maxwell_filter() with st_overlap=True could fail the constant overlap-add (COLA) check when the window had an odd number of samples, and adjust non-COLA windows to the closest COLA window when SciPy 1.16+ is installed, by Eric Larson. (#14312)

• Fix bug with mne.preprocessing.interpolate_bridged_electrodes() where inst.info["bads"] was cleared and not restored when too many electrodes were bridged together and a RuntimeError was raised, by Carina Forster. (#14320)

• Fixed a bug where mne.Epochs.interpolate_bads() raised IndexError for fNIRS data, because the nearest-neighbour donor assignment assumed 2-D (channels, times) data, by Kalle Mäkelä. (#14323)

• Fix mixed-up feature and output axes in multi-output mne.decoding.ReceptiveField inverse patterns, by Daria Agafonova. (#14327)

• Fix bug where mne.io.read_raw_snirf() divided landmark positions by the length unit scale instead of multiplying, placing fiducials far from the head for files not in meters, by Kalle Mäkelä. (#14354)

• Fix bug where Artemis123 axial gradiometers were treated as magnetometers when building SSS external bases, which made the head position estimated by mne.io.read_raw_artemis123() (add_head_trans=True) noisier, by Eric Larson. (#14358)

• Fix bug where hovering over a figure containing several butterfly plots, e.g. from repeated calls to mne.Evoked.plot() or mne.time_frequency.Spectrum.plot() with axes, redrew the entire figure once per plot on every mouse movement, by Eric Larson. (#14368)

• Fix mne.read_evokeds() failing on files that store channel information in the evoked block (as written by MNE-C), by Christoph Dinh. (#14372)

• Fix bug where proj="reconstruct" in mne.Evoked.plot(), mne.Evoked.plot_topomap(), and mne.Evoked.plot_joint() ignored the sphere origin, by Eric Larson. (#14378)

• Fix bug in covariance algebra where SSP projections were checked against the first covariance instead of the second covariance, by Ishaan Maheshwari. (#14387)

• Fix bug where mne.scale_mri() wrote an invalid n_params value to the MRI scaling parameters.cfg file when scale was a scalar, by Christian Brodbeck. (#14389)

• Include pupil channels in baseline correction in Epochs, by Aditya Pandey. (#14390)

• Fix mne.Label.split() dropping vertex 0 with parts="contiguous" and failing on labels without a name, by Christoph Dinh. (#14395)

• Fix bug where mne.io.read_raw_eeglab() and mne.read_epochs_eeglab() raised a false warning when setting montage with EOG channels, by Stefan Appelhoff. (#14396)

• mne.read_epochs() and mne.Epochs.save() no longer warn about BIDS names such as sub-01_task-rest_meg.fif, by Hamza Abdelhedi. (#14398)

• Fix bugs with mne.gui.coregistration() where typing in the scaled subject name field entered the text in reverse, and saving a scaled subject could crash the GUI, by Eric Larson. (#14412)

• Fix Calibration to accept array-like inputs, by Ishaan Maheshwari. (#14413)

• Fix mne.io.Raw.interpolate_bads() for fNIRS data copying an arbitrary channel into bad channels when channel positions are missing, by Kalle Mäkelä. (#14416)

• Fix bug with mne.preprocessing.annotate_muscle_zscore() where channels in raw.info["bads"] were included in the computation of the muscle artifact scores, by Hamza Abdelhedi. (#14418)

• Fix MNE-Python readers for Neuralynx recordings with gaps, following changes in Neo 0.14.6 by Scott Huberty (#14423)

• Handle non-UTF-8 output gracefully in mne.utils.run_subprocess, by Ishaan Maheshwari. (#14425)

• Fix split_naming="bids" since mne.io.Raw.save() and mne.Epochs.save() put split after desc, but split should come before desc, by Hamza Abdelhedi. (#14434)

• Allow passing timezone-aware non-UTC datetime.datetime objects to mne.Info.set_meas_date() and automatically set info["utc_offset"], by Tziona NessAiver and Ishaan Maheshwari. (#14436)

• Fix mne.io.Raw.crop() with reset_first_samp=True misaligning annotations with the data, reading wrong data from files not preloaded, and setting a wrong first_samp on concatenated raws, by Mathieu Scheltienne. (#14439)

• Fix bug where mne.Epochs and mne.EpochsArray accepted an event_id mapping several names to the same event code and only raised an error on data loading or indexing, they now raise at construction, by Stefan Appelhoff. (#14445)

• Fix mne.preprocessing.ICA.get_sources() assigning the wrong channel names and types to add_channels that are not in the order of the instance’s channels, by Stefan Appelhoff. (#14452)

• Fix bug in mne.preprocessing.regress_artifact() where passing betas caused the picks_artifact value to be passed as exclude and the default picks_artifact="eog" to be used instead, by Johannes Herforth. (#14457)

• Fixed several issues with readers and interpolations, by Eric Larson:

  • mne.io.read_raw_eximia() now returns EEG and EOG data in V instead of µV (values are now 1e-6 times what they were before)

  • mne.io.Raw.interpolate_bads() with fNIRS channels no longer clear all of info["bads"] even with reset_bads=False

  • reset_bads=True now removes only the channels that were actually interpolated from info["bads"] instead of all non-excluded bads

  • mne.io.Raw.set_montage() stores EEG digitization points in channel order (with the reference first as ident 0) instead of montage order (#14461)

• Use the smooth head surface packaged with MNE-Python as fsaverage-head.fif in mne.datasets.fetch_fsaverage() (and fix it for a subjects_dir given as a str that does not exist yet), and prefer <subject>-head.fif over outer_skin.surf as the head surface in mne.viz.plot_alignment() and the coregistration GUI, by Christian Brodbeck. (#14462)

• Fix bugs in exact and one-tailed 1-sample permutation tests: n_permutations="all" in mne.stats.permutation_cluster_1samp_test(), mne.stats.spatio_temporal_cluster_1samp_test(), and mne.stats.permutation_t_test() omitted one sign flip, one-tailed 1-sample cluster tests could use the identity twice instead of the global sign flip, and one-tailed mne.stats.permutation_t_test() returned two-tailed p-values. One-tailed mne.stats.permutation_t_test() p-values are now typically about half as large, and other affected p-values can shift by one permutation, by Eric Larson and Cora Kim. (#14482)

• Avoid duplicate extension when saving source estimates with hemisphere tags, and support ftype='auto' to infer format in mne.SourceEstimate.save() and mne.VolSourceEstimate.save(), by Shresth Keshari and Ishaan Maheshwari. (#14491)

• Fix mne.transforms.rot_to_quat() returning wrong quaternions for some rotations near 180°, which made some MEG sensors appear rotated incorrectly in 3D plots, by Eric Larson. (#14492)

• Reduce memory consumption and improve execution speed in mne.stats.permutation_t_test() by chunking permutation evaluations, by Ishaan Maheshwari. (#14494)

• Fix bug where saving annotations with orig_time=None to CSV files corrupted orig_time and shifted onset when loaded with mne.read_annotations(), by Ishaan Maheshwari. (#14503)

• Fix reading BrainVision files whose [Comment] section contains a Channels table without filter columns, by Clemens Brunner. (#14512)

• Fix mne.io.read_raw_brainvision() classifying channels as misc when their unit is written with a Greek mu (μV), by Clemens Brunner. (#14513)

API changes by deprecation

• mne.datasets.lite_data.data_path() is deprecated and will be removed in MNE 1.15; the documentation build now serves the browser notebooks from the regular datasets, by Eric Larson. (#14157)

New features

• Add new API for cluster permutation statistics of sensor data: mne.stats.cluster_test(), see the tutorial Group-level cluster permutation testing with formula contrasts, by Carina Forster, Sophie Herbst, new contributor Maximilien Chaumon, and Scott Huberty. (#12663)

• Speed up mne.baseline.rescale() when channel picks are specified by computing baseline statistics only on the selected channels, by new contributor Ishaan Maheshwari. (#14377)

• Show rejection and flat thresholds in mne.Report.add_epochs(), by new contributor Sruthi Sureshkumar. (#14448)

• Add on_drop_all parameter to mne.Epochs.drop() to control behavior when all epochs are dropped, by Shruti Bhale and Stefan Appelhoff. (#13556)

• Add mne.setup_subcortical_source_space() to build a SourceSpaces from a triangulated surface mesh of a subcortical or cerebellar structure, either from a label or externally-produced mesh surface, by Payam Sadeghi-Shabestari. (#14130)

• Add an “Open in JupyterLite” badge to the tutorials and examples that can run in the browser, by Natneal B. (#14157)

• The mne.viz.Brain trace list now shows picked labels with a short name, their number of source vertices, and the extraction mode, and the traces plot now rescales its y-axis whenever a new trace is picked, by Payam Sadeghi-Shabestari. (#14270)

• mne.viz.Brain now offers 'flat' surf option for subjects that have the required patch files, by Payam Sadeghi-Shabestari. (#14308)

• Add set_head_sphere() method to Raw, Epochs and Evoked to store the head sphere in the .info dictionary instead of having to specify the sphere parameter for every figure that draws some form of topomap, by Marijn van Vliet (#14313)

• Add parameter source_cov for mne.minimum_norm.make_inverse_operator(), allowing use of a custom (diagonal) source covariance matrix for the inverse solution, by Teemu Taivainen. (#14315)

• IDEs that display docstrings through static analysis (e.g., VS Code) now show complete docstrings for MNE-Python functions, methods, and classes, by Eric Larson. (#14316)

• Add Array API support to mne.decoding.ReceptiveField with compatible scikit-learn estimators, preserving the input backend for predictions, scores, and patterns, by Daria Agafonova. (#14327)

• Exporting 2 GB or more of data to EEGLAB with mne.export.export_raw() or mne.export.export_epochs() now uses the MATLAB v7.3 (HDF5) format (requires eeglabio >= 0.1.4 and h5py), by Eric Larson. (#14338)

• Add a theme parameter to mne.gui.coregistration(), and use a dark default 3D background with readable dock titles, legend, and orientation axes, by Payam Sadeghi-Shabestari. (#14353)

• Speed up mne.preprocessing.maxwell_filter() with head movement compensation by avoiding repeated computations when computing the SSS basis, by Eric Larson. (#14358)

• Add timing metadata to GIFTI files written by mne.SourceEstimate.save_as_surface(), allowing readers to recover the original time axis, by Peter Molfese. (#14361)

• Add selectable parameter to mne.time_frequency.Spectrum.plot() and mne.time_frequency.EpochsSpectrum.plot() to disable interactive features, by Eric Larson. (#14368)

• mne.preprocessing.ICA gained a converged_ attribute recording whether the decomposition converged or stopped at max_iter, for all four backends. It is written to and read from -ica.fif files, so a truncated decomposition can be identified after the fact rather than only from a warning at fit time. mne.preprocessing.infomax() gained a matching return_converged parameter, by Aslan Satary Dizaji. (#14370)

• Add mne.channels.DigMontage.drop_channels() to drop channels from a montage, by Stefan Appelhoff. (#14397)

• Add a toolbar (screenshot, show/hide controls, help) with keyboard shortcuts to mne.gui.coregistration(), and show ICP fitting progress with a button (or Esc) to stop the fit early, by Payam Sadeghi-Shabestari. (#14412)

• Add component pagination and temporal sampling sliders for ICA source timecourses in mne.Report.add_ica(), by Stefan Appelhoff. (#14422)

• Add support for higher-resolution image exports via a scale parameter in mne.viz.Brain.screenshot() and mne.viz.Brain.save_image(), by Ishaan Maheshwari. (#14447)

• Add mne.channels.compute_interpolation_data() and support for passing precomputed interpolation data in interpolate_bads(), interpolate_bads(), and interpolate_bads(), by Ishaan Maheshwari. (#14455)

• Store parent-surface vertex numbers in the GIFTI files written by mne.SourceEstimate.save_as_surface(), which now raises an error instead of writing misaligned data when the source estimate and source space vertices differ, by Peter Molfese. (#14510)

• Add support for binary file-like objects (such as io.BytesIO) in mne.io.BaseRaw.save(), by Bruno Aristimunha. (#14516)

• Speed up mne.Annotations.copy(), mne.Annotations.crop(), and mne.io.Raw.set_annotations() by about 10x for thousands of annotations, by Eric Larson. (#14518)

• Add support for continuous data segmentation in mne.time_frequency.psd_array_multitaper() and mne.io.Raw.compute_psd() via n_per_seg, n_overlap, and average parameters, by Ishaan Maheshwari. (#14519)

Other changes

• Added a relevant reference to Frequency and time-frequency sensor analysis, by new contributor Isaac Kinley. (#14304)

• Optimize time slicing order in mne.Epochs.get_data() to avoid temporary full-array copies when subsetting, by new contributor Ishaan Maheshwari. (#14384)

• Optimize multidimensional indexing in mne.Epochs.get_data() to eliminate intermediate 3D array allocations when subsetting both epochs and channels, by new contributor Ishaan Maheshwari. (#14391)

• Optimize multidimensional indexing in mne.time_frequency.Spectrum.get_data() and mne.time_frequency.EpochsTFR.get_data() to eliminate intermediate array allocations, by new contributor Ishaan Maheshwari. (#14402)

• Fixed a typo in the filtering background tutorial, by new contributor Xuzhou Qin. (#14487)

• Clarify the layout of the MEG coil orientation matrix in info["chs"][i]["loc"] for multi-axis OPMs in mne.Info, by new contributor Thais Marchetti. (#14522)

• Add the setup cell that installs MNE into the browser kernel for the JupyterLite documentation, by Natneal B. (#14150)

• Several usability improvements to mne.gui.coregistration(): icons on more action buttons and polished tooltip text, a color-coded fit-quality indicator next to the HSP/MRI distance summary, by Payam Sadeghi-Shabestari. (#14285)

• Lock the JupyterLite browser kernel’s extra packages at build time instead of installing them from the first notebook cell, by Teon Brooks. (#14314)

• Name the offending channels in warnings from mne.preprocessing.nirs.optical_density() and mne.preprocessing.nirs.temporal_derivative_distribution_repair(), by Kalle Mäkelä. (#14326)

• Move the interactive time-viewer code of mne.viz.Brain into a separate private module by Payam Sadeghi-Shabestari. (#14359)

• Vectorize blink interval masking in mne.preprocessing.eyetracking.interpolate_blinks(), by Ishaan Maheshwari. (#14406)

• Document how mne.preprocessing.annotate_muscle_zscore() combines channels into its score and why artifacts on only a few channels can be missed, by Hamza Abdelhedi. (#14419)

• Accelerate Morlet cross-spectral density calculation across channel pairs in mne.time_frequency.csd_morlet() and mne.time_frequency.csd_array_morlet() via batched matrix multiplication, by Ishaan Maheshwari. (#14449)

• Accelerate electrical distance computation in mne.preprocessing.compute_bridged_electrodes() via vectorized sample covariance, by Ishaan Maheshwari. (#14451)