ICA-sEMG

sEMG signals contain cross talk, and a blind source separation algorithm is needed to identify the source signals.

We use the ICA algorithm[1] to separate a set of sEMG source signals from a set of mixed sEMG signals, the implementation of the ICA algorithm is done using python-picard[2-3]

How To Use

ica_analysis includes:

• identification of the source signals
• time plot of the source signals (with the option to scroll horizontally and/or include annotations)
• heat maps of the source signals
• spectral density plots of the source signals

To use ica_analysis you need to convert your data to a numpy array, and import the script using:

from ica_analysis import ica

Parameter Documentation

ica_analysis.ica(sigbufs, sampling_rate, heatmap_bool=False, image_path=None, hscroll_bool=False, annotations_bool=False, set_annotations=None, set_times=None)

• sigbufs (numpy array): signal data such that each row contains the data from a specific channel
• sampling_rate (int): number of samples per second
• heatmap_bool (bool, optional): boolean variable indicating whether you'd like to have the heatmap displayed (default is False)
• image_path (string, optional): path of the image location for the heatmaps (default is None)
• hscroll_bool (bool, optional): boolean variable indicating whether you'd like to have the horizontal scroll in the time plot displayed (default is False)
• annotations_bool (bool, optional): boolean variable indicating whether you'd like to have the annotations in the time plot displayed (default is False)
• set_annotations (numpy array, optional): annotation array (default is None)
• set_times (numpy array, optional): annotation times array (default is None)

sEMG Data Processing

Additionally the repo contains scripts to process sEMG data in two formats:

• EDF: has the option to split the file to multiple, smaller files. The number of files is determined by "num_sets". Additionally user must fill out the path where the EDF is stored ("edf_path")
• CSV: user must fill out the path of the CSV file

Required Python Libraries

• picard
• scipy
• numpy
• matplotlib >= 3.4.2
• math
• cv2 (for selecting electrode location)
• pandas

References

[1] A. Hyvärinen, E. Oja, Independent component analysis: algorithms and applications, Neural Networks, Volume 13, Issues 4–5, 2000, Pages 411-430, ISSN 0893-6080, https://doi.org/10.1016/S0893-6080(00)00026-5.

[2] Pierre Ablin, Jean-Francois Cardoso, Alexandre Gramfort Faster independent component analysis by preconditioning with Hessian approximations IEEE Transactions on Signal Processing, 2018 https://arxiv.org/abs/1706.08171

[3] Pierre Ablin, Jean-François Cardoso, Alexandre Gramfort Faster ICA under orthogonal constraint ICASSP, 2018 https://arxiv.org/abs/1711.10873