This repository contains a set of functions to pre-process and process electroencephalography (EEG) data.

With most recording devices, EEG data are structured as a big matrix of shape (time x electrodes). One electrode channel generaly corresponds to the trigger channel used to synchronise the participant response or the stimuli to the EEG signal. The raw EEG can be split in chunks of time according to this trigger channel. It is then possible to average EEG signal coming from same condition for instance.

These functions can be used to load data, do some kind of processing, plot etc.

This denoising method is an implementation of this matlab toolbox created by Alain de Cheveigné. More details about this method can be found in the following papers:

Plot all electrodes with an offset from t0 to t1. The stimulus channel is also ploted and red lines are used to show the events.

• datainstance of pandas.core.DataFrame

  Add offset to data.

• offsetfloat

  Value of the offset.

Returns:

• newDatainstance of pandas.core.DataFrame

  The data with offset applied to each electrode.

Calculate and return the baseline (average of each data point) of a signal. The baseline will calculated from the first baselineDur seconds of this signal.

• datainstance of pandas.core.DataFrame

  Data used to calculate the baseline.

• baselineDurfloat

  Duration of the baseline to use for the calulation of the average in seconds.

• fsfloat

  Sampling frequency of data in Hz.

Returns:

• baselinefloat

  The baseline value.

Design a filter with scipy functions avoiding unstable results (when using ab output and filtfilt(), lfilter()...). Cf. ()[]

• datainstance of numpy.array | instance of pandas.core.DataFrame

  Data to be filtered. Each column will be filtered if data is a dataframe.

• cutoffarray-like of float

  Pass and stop frequencies in order:

  • the first element is the stop limit in the lower bound
  • the second element is the lower bound of the pass-band
  • the third element is the upper bound of the pass-band
  • the fourth element is the stop limit in the upper bound For instance, [0.9, 1, 45, 48] will create a band-pass filter between 1 Hz and 45 Hz.

• gstopint

  The minimum attenuation in the stopband (dB).

• gpassint

  The maximum loss in the passband (dB).

Returns:

• filteredDatainstance of numpy.array | instance of pandas.core.DataFrame

  The filtered data.

Check filtering and downsampling by ploting both datasets.

• data1instance of pandas.core.DataFrame

  First dataframe.

• data2instance of pandas.core.DataFrame

  Second dataframe.

• fs1float

  Sampling frequency of the first dataframe in Hz.

• fs2float

  Sampling frequency of the second dataframe in Hz.

• startfloat

  Start of data to plot in seconds.

• endfloat

  End of data to plot in seconds.

• electrodeNumint

  Index of the column to plot.

Returns:

• figinstance of matplotlib.figure.Figure

  The figure containing both dataset plots.

Check filtering and downsampling by ploting both datasets.

• data1instance of pandas.core.DataFrame

  First dataframe.

• data2instance of pandas.core.DataFrame

  Second dataframe.

• fs1float

  Sampling frequency of the first dataframe in Hz.

• fs2float

  Sampling frequency of the second dataframe in Hz.

• startfloat

  Start of data to plot in seconds.

• endfloat

  End of data to plot in seconds.

• electrodeNumint

  Index of the column to plot.

Returns:

• figinstance of matplotlib.figure.Figure

  The figure containing both dataset plots.

Compute the FFT of data and return also the axis in Hz for further plot.

• dataarray

  First dataframe.

• fsfloat

  Sampling frequency in Hz.

Returns:

• fAxinstance of numpy.array

  Axis in Hz to plot the FFT.

• fftDatainstance of numpy.array

  Value of the fft.

Calculate the relative energy at the frequency pickFreq from the the FFT of data. Compare the mean around the pick with the mean of a broader zone for each column.

• dataarray-like

  Matrix of the shape (time, electrode).

• pickFreqfloat

  Frequency in Hz of the pick for which we want to calculate the relative energy.

• showPlotboolean

  A plot of the FFT can be shown.

• fsfloat

  Sampling frequency in Hz.

Returns:

• pickRatiofloat

  Relative energy of the pick.

Create stim channel from events.

• eventsinstance of pandas.core.DataFrame

  Dataframe containing list of events obtained with mne.find_events(raw) .

Returns:

• stiminstance of pandas.core.series.Series

  Series containing the stimulus channel reconstructed from events.

Discriminate triggers when different kind of events are on the same channel. A time threshold is used to determine if two events are from the same trial.

• eventsinstance of pandas.core.DataFrame

  Dataframe containing the list of events obtained with mne.find_events(raw).

• thresholdfloat

  Time threshold in milliseconds. Keeps an event if the time difference with the next one is superior than threshold.

Returns:

• newDatainstance of pandas.series.Series

  List of trial number filling the requirements.

Resample data from oldFS to newFS using the scipy 'resample' function.

• datainstance of pandas.core.DataFrame

  Data to resample.

• oldFSfloat

  The sampling frequency of data.

• newFSfloat

  The new sampling frequency.

Returns:

• newDatainstance of pandas.DataFrame

  The downsampled dataset.

Modify the timestamps of events to match a new sampling frequency.

• eventsinstance of pandas.core.DataFrame

  Dataframe containing list of events obtained with mne.find_events(raw) .

• oldFSfloat

  The sampling frequency of the input events.

• newFSfloat

  The sampling frequency to the output events.

Returns:

• newEventsinstance of pandas.DataFrame

  DataFrame containing the downsampled events.

Resample data from oldFS to newFS using the scipy 'resample' function.

• datainstance of pandas.core.DataFrame

  Data to resample.

• oldFSfloat

  The sampling frequency of data.

• newFSfloat

  The new sampling frequency.

Returns:

• newDatainstance of pandas.DataFrame

  The downsampled dataset.

Fetch behavior data from couchdb (SOA, SNR and trial duration).

• dbAddressstr

  Path to the couch database.

• dbNamestr

  Name of the database on the couch instance.

• sessionNumint

  Behavior data will be fetched from this sessionNum.

Returns:

• lookupTableinstance of pandas.core.DataFrame

  A dataframe containing trial data.

Get the events corresponding to eventCode.

• rawinstance of mne.io.edf.edf.RawEDF

  RawEDF object from the MNE library containing data from the .bdf files.

• eventCodeint

  Code corresponding to a specific events. For instance, with a biosemi device, the triggers are coded 65284, 65288 and 65296 respectively on the first, second and third channel.

Returns:

• startEventsinstance of pandas.core.DataFrame

  Dataframe containing the list of timing corresponding to the event code in the first column. The second column contains the code before the event and the third the code of the selected event.

Returns a subset of table according to SOA, SNR and/or targetFreq. This is used to select trials with specific parameters.

• tableinstance of pandas.core.DataFrame

  DataFrame containing trial number and their parameters (SOA, SNR...).

• kwargsarray-like of int | None

  Array containing element from table to select. It can be SOA, SNR or targetFreq.

Returns:

• newDatainstance of pandas.series.Series

  List of trial number filling the requirements.

Load data from .bdf files. If an array of path is provided, files will be concatenated.

• pathstr | array-like of str

  Path to the .bdf file(s) to load.

Returns:

• rawinstance of mne.io.edf.edf.RawEDF

  RawEDF object from the MNE library containing data from the .bdf files.

Normalize data by subtracting the baseline to each data point. The data used to normalize has to be included at the beginning of data. For instance, to normalize a 10 seconds signal with a 0.1 second baseline, data has to be 10.1 seconds and the baseline used will be the first 0.1 second.

• datainstance of pandas.core.DataFrame

  Data to normalize.

• baselineDurfloat

  Duration of the baseline to use for the normalization in seconds.

• fsfloat

  Sampling frequency of data in Hz.

Returns:

• normalizedinstance of pandas.core.DataFrame

  The normalized data.

Plot all electrodes with an offset from t0 to t1. The stimulus channel is also ploted and red lines are used to show the events.

• datainstance of pandas.core.DataFrame

  Data to plot (not epoched). Columns correspond to electrodes.

• stiminstance of pandas.core.DataFrame

  One column dataframe containing the event codes. Used to plot the stimulus timing along with EEG.

• eventsinstance of pandas.core.DataFrame

  Dataframe containing the list of events obtained with mne.find_events(raw).

• offsetfloat

  Offset between each electrode line on the plot.

• t0float

  Start of data to plot.

• t1float

  End of data to plot.

• fsfloat

  Sampling frequency of data in Hz.

Returns:

• figinstance of matplotlib.figure.Figure

  The figure of the data subset in the time domain.

startOffset=0):

freqMin=None, freqMax=None, yMin=None, yMax=None, startOffset=0, noiseAve=None):

Plot the filter frequency response.

• zpkarray-like

  The 3 parameters of the filter [z, p, k].

• fsfloat

  Sampling frequency in Hz.

Returns:

• figinstance of matplotlib.figure.Figure

  The figure of the filter response.

Transform each electrode of data according to the average of M1 and M2.

• datainstance of pandas.core.DataFrame

  First column has to contain the timing of events in frames.

• M1instance of pandas.core.series.Series

  Values of mastoid 1. This Series has to be the same length as data.

• M2instance of pandas.core.series.Series

  Values of mastoid 2. This Series has to be the same length as data

Returns:

• newDatainstance of pandas.core.DataFrame

  A dataframe referenced to matoids containing all electrode from which we subtract the average of M1 and M2.

It uses some methods of the MNE library and heavily depends on pandas and numpy.