In ymatts/emgR: surface electromyography (sEMG) analyzer with R

emgR overview

• Analysis suites for surface electromyography (sEMG) data
• Basic process functionality
  • Bandpass filter
  • Qunatification
  • Normalization
• Advanced analysis
  • Waveform simililarity analysis
    • Coherence analysis (incl. wavelet)
    • Auto/cross correlation analysis
  • Similiarity based muscle network analysis

Install emgR package

# install.packages("devtools") if not installed
 devtools::install_github("https://github.com/ymatts/emgR")
#install.packages("~/Dropbox/matsui/git/emgR.tar.gz",repos = NULL,type = "source")
library(emgR)

Data acquisition

• From Santuz et al. (2018)
• 13 muscles sEMG (1000 Hz) records during gait cycle
• Time data will be required for constructing emgR object

data(emg)
head(emg,3)

Constructing object

• emgList is a method for constructing emgR object
• emgList object is simply a list data structure
• Input time (sec) and multi channel sEMG data
  • Time is in row and Muscles are in colums

tt = emg$Time # Time record
xx = emg[,-1] # EMG record
obj = emgR::emgList(x = xx,t = tt)
obj

Add annotation

• Annotation file must be following format
  • id in first column may indicate one cycle or trial
  • type in second column may indicate description of id
  • start and end should be annotated/triggered time in second

(f = system.file("extdata","gaitAnno.txt",package = "emgR"))
head(read.table(f,header = T,sep="\t"),3)

Add annotation

• addAnnotation function adds indices in raw emg records with annotated time

obj = emgR::addAnnotation(obj,annoFile = f)
head(obj$annotation,3)

Extract EMG signals followed by annotation

Filtering sEMG signals

• Bandpass filter using butterworth filter
• Filtered objects will be overwride in emgR
• Default setting is order 4 and [5,200] bandpass filter

obj = emgR::btw(obj,order = 4,filtFreq = c(5,200))
obj

Normalize sEMG signals

• Several normalizing methods will be implemted
  • Default setting use simple Maximum Voluntary Contruct (MVC)

obj = emgR::normalize(obj)
obj

Extract gait cycle

• 41 gait cycles
• Each line corresponds to a cycle
  • First column is times during heel contact
  • Second column is times during stance

load("data/cycleTime.rda")
head(cycleTime)

Waveform similiarity analysis

• Coherence analysis
  • Linear coherence / Wavelet coherence
• Correlation analysis
  • Auto/Cross-correlation, Distance correlation etc.
• Default setting is linear coherence by stats::spec.pgram() function

obj = emgR::coherence(obj)
obj

Visualize coherence function

• Muscle $^times$ Muscle smoothed coherences are plotted
• The plot is saved in file path with files
• cohRange is which frequency range should be considered
• This takes several time

file_path = tempfile()
plot(obj,type="coh",files = file_path,cohRange=c(0,100))

Visualize coherence functions

knitr::include_graphics(file_path)

Aggregate coherence function into means within frequency bands

• Average aggregation by frequecies such as $\alpha$, $\beta$, $\gamma$ bands.
• Frequency can be specified with cohRange by vector.
  • Following example shows averaged coherence from 15 Hz to 20 Hz.
• Heatmap of aggregated coherence can be visulized by plot.it argument

obj = emgR::aggrCoh(obj,cohRange = c(15,20),plot.it = FALSE)
obj

Aggregate coherence function into means within frequency bands

obj = emgR::aggrCoh(obj,cohRange = c(15,20),plot.it = TRUE)

Coherent muscle network construction

• Network is constructed via aggregated coherence
• Two mode of networks:
  • Weighted network can be constructed with mode="weighted"
  • Not weighted network can be constructed with mode="bin"
• Coherence is binarized with thres argument
  • Default thres is 0.25
• You can visualize network via plot.it argument.

obj = emgR::cohNet(obj,mode = "bin",thres = 0.25,plot.it = FALSE)

Coherent muscle network construction

obj = emgR::cohNet(obj,mode = "bin",thres = 0.25,plot.it = TRUE)

Network based muscle module detection

• Muscle modules are found based on linkcomm pakcage (Ahn et al. 2010, Becker et al. 2012)
• plot.it visualize module structure
  • You can choose plot.type from "member" or "graph"
• In this model, each muscle can belong to several modules (muti-membership)

Network based muscle module detection

• In case of plot.type = "graph"

obj = emgR::moduleFinder(obj,plot.it = TRUE,plot.type = "graph")

Network based muscle module detection

• In case of plot.type = "member"

obj = emgR::moduleFinder(obj,plot.it = TRUE,plot.type = "member")

Muscle synergy (2D-array)

• Derive muscle synergy based on Non-negative matrix factorzation (NMF) algorithm

ymatts/emgR documentation built on March 7, 2020, 12:52 a.m.