MNE-compatible package for SSVEP analysis
Clone or download

README.md

ssvepy

A package to analyse MNE-formatted EEG data for steady-state visually evoked potentials (SSVEPs).

Install:

pip install git+https://github.com/janfreyberg/ssvepy.git

As always with pip packages, you can install a "development" version of this package by (forking and) cloning the git repository and installing it via pip install -e /path/to/package. Please do open a pull request if you make improvements.

Documentation:

The docs for this package are at http://www.janfreyberg.com/ssvepy. There, you'll find the API and an example notebook.

Usage:

You should load, preprocess and epoch your data using MNE.

Take a look at a notebook that sets up an SSVEP analysis structure with the example data in this package: https://github.com/janfreyberg/ssvepy/blob/master/example.ipynb

Once you have a data structure of the class Epoch, you can use ssvepy.Ssvep(epoch_data, stimulation_frequency), where stimulation_frequency is the frequency (or list of frequencies) at which you stimulated your participants.

Other input parameters and their defaults are:

  • The following parameters, which are equivalent to the parameters in mne.time_frequency.psd_multitaper:
    • fmin=0.1, the low end of the frequency range
    • fmax=50, the high end of the frequency range
    • tmin=None, the start time of the segment you want to analyse
    • tmax=None, the end time of the segment you want to analyse
  • noisebandwidth=1.0, what bandwidth around a frequency should be used to calculate its signal-to-noise-ratio
  • Whether you want to compute the following nonlinearity frequencies:
  • compute_harmonics=True
  • compute_subharmonics=False
  • compute_intermodulation=True (NB: only when there's more than one input frequency)
  • You can also provide your own Power-spectrum data, if you have worked it out using another method.
    • psd=None The powerspectrum. Needs to be a numpy array with dimensions: (epochs, channels, frequency)
    • freqs=None The frequencys at which the powerspectrum was evaluated. Needs to be a one-dimensional numpy array.

The resulting data has the following attributes:

  • stimulation: a data structure with the following attributes:
    • stimulation.frequencies, stimulation.power, stimulation.snr
  • harmonics, subharmonics, intermodulations: non-linear combination of your input stimulus frequencies, all with the attributes:
    • _.frequencies, _.power, _.snr, _.order
  • psd: the Power-spectrum
  • freqs: the frequencies at which the psd was evaluated

And the following methods:

  • plot_psd(): Plot the power spectrum
  • plot_snr(): Plot the SNR spectrum
  • save(filename): Saves an hdf5 file that can be loaded with ssvepy.load_ssvep(filename) 1

More to come.


1: This package currently uses hierarchical data files (hdf5) because it seems to lend itself to the different data stored in ssvep classes, but I know it's less than ideal to have different data structures from MNE. I'm still thinking about improvements.