# Importing data into MNE¶

This guide covers how to import data into MNE python. It includes instructions for importing from common recording equipment in MEG and EEG, as well as how to import raw data from numpy arrays.

## Importing MEG data¶

This section describes the data reading and conversion utilities included with the MNE software. The cheatsheet below summarizes the different file formats supported by MNE software.

Datatype File format Extension MNE-Python function
MEG 4-D Neuroimaging / BTI dir mne.io.read_raw_bti()
EEG European data format .edf mne.io.read_raw_edf()
EEG Biosemi data format .bdf mne.io.read_raw_edf()
EEG General data format .gdf mne.io.read_raw_edf()
EEG EGI simple binary .egi mne.io.read_raw_egi()
EEG EGI MFF format .mff mne.io.read_raw_egi()
Electrode locations elc, txt, csd, sfp, htps Misc mne.channels.read_montage()
Electrode locations EEGLAB loc, locs, eloc Misc mne.channels.read_montage()

Note

All IO functions in MNE-Python performing reading/conversion of MEG and EEG data can be found in mne.io and start with read_raw_. All supported data formats can be read in MNE-Python directly without first saving it to fif.

Note

Irrespective of the units used in your manufacturer’s format, MNE-Python will always use the units listed below and perform conversions during the IO procedure if necessary.

• V: eeg, eog, seeg, emg, ecg, bio, ecog
• T: mag
• M: hbo, hbr
• Am: dipole
• AU: misc

Note

MNE-Python performs all computation in memory using the double-precision 64-bit floating point format. This means that the data is typecasted into float64 format as soon as it is read into memory. The reason for this is that operations such as filtering, preprocessing etc. are more accurate when using the double-precision format. However, for backward compatibility, it writes the fif files in a 32-bit format by default. This is advantageous when saving data to disk as it consumes less space.

However, if the users save intermediate results to disk, they should be aware that this may lead to loss in precision. The reason is that writing to disk is 32-bit by default and then typecasting to 64-bit does not recover the lost precision. In case you would like to retain the 64-bit accuracy, there are two possibilities:

• Chain the operations in memory and not save intermediate results
• Save intermediate results but change the dtype used for saving. However, this may render the files unreadable in other software packages

### Elekta NeuroMag (.fif)¶

Note

If the data were recorded with MaxShield on and have not been processed with MaxFilter, they may need to be loaded with mne.io.read_raw_fif(..., allow_maxshield=True).

### Importing 4-D Neuroimaging / BTI data¶

MNE-Python includes the mne.io.read_raw_bti() to read and convert 4D / BTI data. This reader function will by default replace the original channel names, typically composed of the letter A and the channel number with Neuromag. To import the data, the following input files are mandatory:

• A data file (typically c,rfDC) containing the recorded MEG time-series.
• A hs_file containing the digitizer data.
• A config file containing acquisition information and metadata.

By default mne.io.read_raw_bti() assumes these three files to be located in the same folder.

Note

While reading the reference or compensation channels, currently, the compensation weights are not processed. As a result, the mne.io.Raw object and the corresponding fif file does not include information about the compensation channels and the weights to be applied to realize software gradient compensation. To augment the Magnes fif files with the necessary information, the command line tools include the utilities mne_create_comp_data, and mne_add_to_meas_info. Including the compensation channel data is recommended but not mandatory. If the data are saved in the Magnes system are already compensated, there will be a small error in the forward calculations whose significance has not been evaluated carefully at this time.

#### Creating software gradient compensation data¶

The utility mne_create_comp_data was written to create software gradient compensation weight data for 4D Magnes fif files. This utility takes a text file containing the compensation data as input and writes the corresponding fif file as output. This file can be merged into the fif file containing 4D Magnes data with the utility mne_add_to_meas_info. See mne_create_comp_data for command-line options.

### Importing CTF Polhemus data¶

The CTF MEG systems store the Polhemus digitization data in text files. The utility mne_ctf_dig2fiff was created to convert these data files into the fif and hpts formats.

Since the software gradient compensation employed in CTF systems is a reversible operation, it is possible to change the compensation status of CTF data in the data files as desired. This section contains information about the technical details of the compensation procedure and a description of mne_compensate_data , which is a utility to change the software gradient compensation state in evoked-response data files.

The fif files containing CTF data converted using the utility mne_ctf2fiff contain several compensation matrices which are employed to suppress external disturbances with help of the reference channel data. The reference sensors are located further away from the brain than the helmet sensors and are thus measuring mainly the external disturbances rather than magnetic fields originating in the brain. Most often, a compensation matrix corresponding to a scheme nicknamed Third-order gradient compensation is employed.

Let us assume that the data contain $$n_1$$ MEG sensor channels, $$n_2$$ reference sensor channels, and $$n_3$$ other channels. The data from all channels can be concatenated into a single vector

$x = [x_1^T x_2^T x_3^T]^T\ ,$

where $$x_1$$, $$x_2$$, and $$x_3$$ are the data vectors corresponding to the MEG sensor channels, reference sensor channels, and other channels, respectively. The data before and after compensation, denoted here by $$x_{(0)}$$ and $$x_{(k)}$$, respectively, are related by

$x_{(k)} = M_{(k)} x_{(0)}\ ,$

where the composite compensation matrix is

$\begin{split}M_{(k)} = \begin{bmatrix} I_{n_1} & C_{(k)} & 0 \\ 0 & I_{n_2} & 0 \\ 0 & 0 & I_{n_3} \end{bmatrix}\ .\end{split}$

In the above, $$C_{(k)}$$ is a $$n_1$$ by $$n_2$$ compensation data matrix corresponding to compensation “grade” $$k$$. It is easy to see that

$\begin{split}M_{(k)}^{-1} = \begin{bmatrix} I_{n_1} & -C_{(k)} & 0 \\ 0 & I_{n_2} & 0 \\ 0 & 0 & I_{n_3} \end{bmatrix}\ .\end{split}$

To convert from compensation grade $$k$$ to $$p$$ one can simply multiply the inverse of one compensate compensation matrix by another and apply the product to the data:

$x_{(k)} = M_{(k)} M_{(p)}^{-1} x_{(p)}\ .$

This operation is performed by mne_compensate_data.

### Importing KIT MEG system data¶

MNE-Python includes the mne.io.read_raw_kit() and mne.read_epochs_kit() to read and convert KIT MEG data. This reader function will by default replace the original channel names, which typically with index starting with zero, with ones with an index starting with one.

To import continuous data, only the input .sqd or .con file is needed. For epochs, an Nx3 matrix containing the event number/corresponding trigger value in the third column is needed.

The following input files are optional:

• A KIT marker file (mrk file) or an array-like containing the locations of the HPI coils in the MEG device coordinate system. These data are used together with the elp file to establish the coordinate transformation between the head and device coordinate systems.
• A Polhemus points file (elp file) or an array-like containing the locations of the fiducials and the head-position indicator (HPI) coils. These data are usually given in the Polhemus head coordinate system.
• A Polhemus head shape data file (hsp file) or an array-like containing locations of additional points from the head surface. These points must be given in the same coordinate system as that used for the elp file.

Note

The output fif file will use the Neuromag head coordinate system convention, see The head and device coordinate systems. A coordinate transformation between the Polhemus head coordinates and the Neuromag head coordinates is included.

By default, KIT-157 systems assume the first 157 channels are the MEG channels, the next 3 channels are the reference compensation channels, and channels 160 onwards are designated as miscellaneous input channels (MISC 001, MISC 002, etc.). By default, KIT-208 systems assume the first 208 channels are the MEG channels, the next 16 channels are the reference compensation channels, and channels 224 onwards are designated as miscellaneous input channels (MISC 001, MISC 002, etc.).

In addition, it is possible to synthesize the digital trigger channel (STI 014) from available analog trigger channel data by specifying the following parameters:

• A list of trigger channels (stim) or default triggers with order: ‘<’ | ‘>’ Channel-value correspondence when converting KIT trigger channels to a Neuromag-style stim channel. By default, we assume the first eight miscellaneous channels are trigger channels. For ‘<’, the largest values are assigned to the first channel (little endian; default). For ‘>’, the largest values are assigned to the last channel (big endian). Can also be specified as a list of trigger channel indexes.
• The trigger channel slope (slope) : ‘+’ | ‘-‘ How to interpret values on KIT trigger channels when synthesizing a Neuromag-style stim channel. With ‘+’, a positive slope (low-to-high) is interpreted as an event. With ‘-‘, a negative slope (high-to-low) is interpreted as an event.
• A stimulus threshold (stimthresh) : float The threshold level for accepting voltage changes in KIT trigger channels as a trigger event.

The synthesized trigger channel data value at sample $$k$$ will be:

$s(k) = \sum_{p = 1}^n {t_p(k) 2^{p - 1}}\ ,$

where $$t_p(k)$$ are the thresholded from the input channel data d_p(k):

$\begin{split}t_p(k) = \Bigg\{ \begin{array}{l} 0 \text{ if } d_p(k) \leq t\\ 1 \text{ if } d_p(k) > t \end{array}\ .\end{split}$

The threshold value $$t$$ can be adjusted with the stimthresh parameter, see below.

## Importing EEG data¶

The MNE package includes various functions and utilities for reading EEG data and electrode templates.

### BrainVision (.vhdr, .vmrk, .eeg)¶

The BrainVision file format consists out of three separate files:

1. A text header file (.vhdr) containing meta data
2. A text marker file (.vmrk) containing information about events in the data
3. A binary data file (.eeg) containing the voltage values of the EEG

Both text files are based on the Microsoft Windows INI format consisting of:

• sections marked as [square brackets]
• comments marked as ; comment
• key-value pairs marked as key=value

A documentation for core BrainVision file format is provided by Brain Products. You can view the specification here

BrainVision EEG files can be read in using mne.io.read_raw_brainvision() with the .vhdr header file as an input.

Warning

Renaming BrainVision files can be problematic due to their multifile structure. See this example for an instruction.

### European data format (.edf)¶

EDF (European Data Format) and EDF+ are 16-bit formats.

The EDF+ files may contain an annotation channel which can be used to store trigger information. The Time-stamped Annotation Lists (TALs) on the annotation data can be converted to a trigger channel (STI 014) using an annotation map file which associates an annotation label with a number on the trigger channel.

### Biosemi data format (.bdf)¶

The BDF format is a 24-bit variant of the EDF format used by the EEG systems manufactured by a company called BioSemi. It can also be read in using mne.io.read_raw_edf().

Warning

The data samples in a BDF file are represented in a 3-byte (24-bit) format. Since 3-byte raw data buffers are not presently supported in the fif format these data will be changed to 4-byte integers in the conversion.

### General data format (.gdf)¶

GDF (General Data Format) is a flexible format for biomedical signals, that overcomes some of the limitations of the EDF format. The original specification (GDF v1) includes a binary header, and uses an event table. An updated specification (GDF v2) was released in 2011 and adds fields for additional subject-specific information (gender, age, etc.) and allows storing several physical units and other properties. Both specifications are supported in MNE.

### Neuroscan CNT data format (.cnt)¶

CNT files can be read in using mne.io.read_raw_cnt(). The channel locations can be read from a montage or the file header. If read from the header, the data channels (channels that are not assigned to EOG, ECG, EMG or misc) are fit to a sphere and assigned a z-value accordingly. If a non-data channel does not fit to the sphere, it is assigned a z-value of 0. See The head and device coordinate systems

Warning

Reading channel locations from the file header may be dangerous, as the x_coord and y_coord in ELECTLOC section of the header do not necessarily translate to absolute locations. Furthermore, EEG-electrode locations that do not fit to a sphere will distort the layout when computing the z-values. If you are not sure about the channel locations in the header, use of a montage is encouraged.

### EGI simple binary (.egi)¶

EGI simple binary files can be read in using mne.io.read_raw_egi(). The EGI raw files are simple binary files with a header and can be exported from using the EGI Netstation acquisition software.

### Importing EEG data saved in the Tufts University format¶

The command line utility mne_tufts2fiff was created in collaboration with Phillip Holcomb and Annette Schmid from Tufts University to import their EEG data to the MNE software.

The Tufts EEG data is included in three files:

• The raw data file containing the acquired EEG data. The name of this file ends with the suffix .raw .
• The calibration raw data file. This file contains known calibration signals and is required to bring the data to physical units. The name of this file ends with the suffix c.raw .
• The electrode location information file. The name of this file ends with the suffix .elp .

See the options for the command-line utility mne_tufts2fiff.

### eXimia EEG data¶

EEG data from the Nexstim eXimia system can be read in using the mne.io.read_raw_eximia() function.

## Setting EEG references¶

The preferred method for applying an EEG reference in MNE is mne.set_eeg_reference(), or equivalent instance methods like raw.set_eeg_reference(). By default, an average reference is used. Instead of applying the average reference to the data directly, an average EEG reference projector is created that is applied like any other SSP projection operator.

There are also other functions that can be useful for other referencing operations. See mne.set_bipolar_reference() and mne.add_reference_channels() for more information.

Some EEG formats (EGI, EDF/EDF+, BDF) neither contain electrode location information nor head shape digitization information. Therefore, this information has to be provided separately. For that purpose all readers have a montage parameter to read locations from standard electrode templates or a polhemus digitizer file. This can also be done post-hoc using the mne.io.Raw.set_montage() method of the Raw object in memory.

When using the locations of the fiducial points the digitization data are converted to the MEG head coordinate system employed in the MNE software, see The head and device coordinate systems.

## Creating MNE data structures from arbitrary data (from memory)¶

Arbitrary (e.g., simulated or manually read in) raw data can be constructed from memory by making use of mne.io.RawArray, mne.EpochsArray or mne.EvokedArray in combination with mne.create_info().

This functionality is illustrated in Creating MNE objects from data arrays. Using 3rd party libraries such as NEO (https://pythonhosted.org/neo/) in combination with these functions abundant electrophysiological file formats can be easily loaded into MNE.

## Importing MEG/EEG data from FieldTrip¶

The data is imported directly from a .mat file.

The info parameter can be explicitly set to None. The import functions will still work but:

1. All channel locations will be in head coordinates.
2. Channel orientations cannot be guaranteed to be accurate.
3. All channel types will be set to generic types.

This is probably fine for anything that does not need that information, but if you intent to do things like interpolation of missing channels, source analysis or look at the RMS pairs of planar gradiometers, you most likely run into problems.

It is highly recommended to provide the info parameter as well. The info dictionary can be extracted by loading the original raw data file with the corresponding MNE-Python functions:

original_info = original_data.info