p0lyis a computational pipeline for EEG data preprocessing and analysis. Its goal is to provide automated and repeatable data clean up and computation of a wide range of analyses. It is written primarily in Python and built atop the MNE M/EEG data processing library and the PyDOIT workflow management system.
The pipeline consists of a set of processing tasks (/scripts), a central configuration file (config.py) and a set of instructions for parsing/computing condition-relevant metadata (event_codes.py). Once the configuration is set up, tasks can be run on datasets with a single command line instruction - without further interaction and with automated execution of all necessary preprocessing steps. After a task completed, a range of Jupyter notebooks can be used to visualize and further investigate the analysis results.
Currently available analyses are Event Related and Global Field Potentials, Time Frequency Analysis (Power and ITPC), Connectivity Analysis (within-frequency and cross-frequency, time-resolved and -averaged), Waveform Geometry Analysis (by-cycle), Parametric and Non-parametric Statistics, Network Analysis, and Single-trial Analysis (Linear Mixed Models).
- Getting started
- Requirements
- Installation
- Remote setup w/ PyCharm
- Project setup
- Data folder
- Config file
- Event codes
- Metadata parsing
####00 - Initialize dataset
This stage scans folder for Brainvision files and assigns subject ids to them, then it create folders for each subject and moves the original files into them. Input: folder with BrainVision files Output: files get uniform subject IDs, moved to subfolders
####01 - Preprocess raw
This stage rereferences the EEG, then detects and interpolates bad channels. After the first round of channel correction, blink and saccade artifacts are removed by subtracting EOG-correlated ICA components. Then a second round of channel correction is applied. Finally, z-score based artifact detection and annotation is performed. The full recording is saved to MNE raw format. In addition, the ICA matrix is saved. Input: BrainVision files and subject ID Output: cleaned and annotated raw files, ICA matrix
####02 - Epoch raw
In this step, the recording is segmented according to the (possibly multiple) lock markers and time intervals defined in the configuration file. If methods for metadata computation are provided they're applied and metadata is attached to the .metadata property. Trials containing segments that have been annotated (as containing artifacts) in the previous step are dropped. The epoched data is saved to MNE epoch format.
Input: cleaned and annotated raw files, configuration with time-locks that are defined by event codes and time intervals
Output: trial-segmented data with trial metadata attached and bad trials removed
This step applies Morlet wavelet convolution to the epoched data and saves the (unaveraged) complex format result in MNE TFR format. The frequencies and width of the wavelets are specified in the freqs and n_cycles variables. csd determines if a spatial filter is applied before the analysis. toi and decim specify the crop and resampling to be applied after wavelet convolution.
Input: trial-segmented data, configuration with frequencies and toi
Output: time-frequency data, cropped and resampled
Analogous to step 3, this step applies windowed FIR bandpass filtering to the epoched data and saves the (unaveraged) complex format result in MNE TFR format. The frequency bands and bandwidths are specified in the freqs and bandwidth variables. toi specifies the crop to be applied after FIRWIN filtering.
Input: trial-segmented data, configuration with frequency bands and toi
Output: time-frequency data, cropped and resampled
####10 - Single-trial Analysis