Hello Mr. Ghaderi,

From reading your question I understand that you have distributed a number of electrodes ( I am assuming surface electrodes and not fine wire) around the wrist and forearm that makes up 12 EMG recording channels. However, from your question it is not obvious if you have immobilised the wrist and the elbow or you have not. Furthermore, I assume the 5 hand movements are repeated a number of times but it is not clear to me if these movements are actual movements or they are isometric contractions and you just refer to them as movements. Therefore I take my assumption from your question literally and assume that these are not isometric contractions. I am now forced to make a further assumption in regards the strengths of the EMG signals from the 12 channels. If you have not used a number of force transducers either to ensure that each time the hand movement is repeated, it is done so with reference to the same force (using some form of biofeedback for the subject) OR you measure the force for each movement and then use it to normalise your EMG signals so that they can be comparable.

So far I have not made any assumptions about your proposed hand movements but if they are the type of movements that your subjects perform in their day-to-day life then your 12 channels may contain different activation patterns because of the differences in the movements despite the synergistic muscle activation that is organised by the Central Nervous System (CNS). However, if your subjects are not familiar with these movements or these movements are not part of the set of movements that your subjects are familiar with and consider normal then you might have different activation patterns in the 12 channels but this time for different reasons. In this scenario the muscle recruitment by the CNS will be inconsistent and because of the built-in redundancy within the musculoskeletal system, the muscle activation pattern may be different each time the same movement is repeated.

You have already had answers as to how to compute the histogram but I doubt that will enable you to classify these signals.

I wish you success.

Hello,

Well, you may simply use “hist” command in MATLAB to display the histogram of your signal. In addition, you may extract several features, say variance, skewness, and kurtosis to describe the profile of sEMG probability density function, quantitatively, being beneficial in automatic machine learning. Here is a couple of MATLAB code:

%% Calculating EMG histogram

figure(1); subplot(211), plot(m1); xlabel('Samples','fontsize',12,'fontweight','b'); title('SCM Flexion','fontsize',12,'fontweight','b'); subplot(212), hist(m1); title('Histogram','fontsize',12,'fontweight','b'); set(gcf, 'color', 'white'); %m1 is the name of data imported in workspace

%%Saving the results in a variable

h=hist(m1);

%% Extracting quantitative features from histograms

%% Cumulants (1-4) for 5-second duration (10000 samples) with sampling frequency of 2kHz

j=1;

for i=60000:2000:68000 % an arbitrary section of my data, which is 5 seconds together with signal framing (subdividing data into 5 frames of 1-sec duration)

c1(j,1) = mean(m1(i+1:i+2000)); %Mean

c2(j,1) = cumest(m1(i+1:i+2000),2); %Variance

c3(j,1) = cumest(m1(i+1:i+2000),3) / c2(j,1)^(3/2); %Skewness

c4(j,1) = cumest(m1(i+1:i+2000),4) / c2(j,1)^2; %Kurtosis

j=j+1;

end

h1=mean(c1);

h2=mean(c2);

h3=mean(c3);

h4=mean(c4);

It, however, necessitates adding HOSA toolbox to calculate the first- to fourth-order cumulants.

Regards,

In case you don't have HOSA toolbox, use...

K1=mean(X) returns the mean value of the elements in X

K2 = var(X) returns the variance of the values in X

K3 = skewness(X) returns the sample skewness of the values in X

K4 = kurtosis(X) returns the sample kurtosis of the values in X.

I am just curious;

which information do you draw from the histogram? Shouldn't it show equal distributions no matter which contraction mode was used without skewness after you removed the signal offset?

And what would your classification look like, what is the criterion for classifying related to which feature of the histogram?

Cheers

Axel

Hello Mr. Ghaderi,

From reading your question I understand that you have distributed a number of electrodes ( I am assuming surface electrodes and not fine wire) around the wrist and forearm that makes up 12 EMG recording channels. However, from your question it is not obvious if you have immobilised the wrist and the elbow or you have not. Furthermore, I assume the 5 hand movements are repeated a number of times but it is not clear to me if these movements are actual movements or they are isometric contractions and you just refer to them as movements. Therefore I take my assumption from your question literally and assume that these are not isometric contractions. I am now forced to make a further assumption in regards the strengths of the EMG signals from the 12 channels. If you have not used a number of force transducers either to ensure that each time the hand movement is repeated, it is done so with reference to the same force (using some form of biofeedback for the subject) OR you measure the force for each movement and then use it to normalise your EMG signals so that they can be comparable.

So far I have not made any assumptions about your proposed hand movements but if they are the type of movements that your subjects perform in their day-to-day life then your 12 channels may contain different activation patterns because of the differences in the movements despite the synergistic muscle activation that is organised by the Central Nervous System (CNS). However, if your subjects are not familiar with these movements or these movements are not part of the set of movements that your subjects are familiar with and consider normal then you might have different activation patterns in the 12 channels but this time for different reasons. In this scenario the muscle recruitment by the CNS will be inconsistent and because of the built-in redundancy within the musculoskeletal system, the muscle activation pattern may be different each time the same movement is repeated.

You have already had answers as to how to compute the histogram but I doubt that will enable you to classify these signals.

I wish you success.

Hello,

I have a question for you that the answer was considered, are the signals free row or row after rectifying (e.g: root mean square), after that you must determine the duration of any movement from the five hand movements (initial and final).

Best regards

Abdel-Rahman Akl : no signals of all channel are free row (without rectifying)

What physiological information do you want to extract from the histograms?

histogram is an extention of ZC(zero crossing) and WAMP (wilson amplitude) measures.both of which compare a single threshold to the EMG signal or a closely related function. Since the EMG signal deviates highly from its baseline when the muscle is in high contraction levels, it follows that it would be informative to measure the frequency with which the EMG signal reaches multiple amplitude levels. A simple way to do this is histogram which by segmenting the signal to predefined bins (level of voltage) do this task. i used " hist" matlab command to do this but unfortunatly a i couldnt get result

You can directly ask Mr. Sadri about your problem. As you know, he knows everything!

I have the same question. Did you find out how to extract this feature?

yes here is the Matlab function:

function out= gethistfeat(data,nums,win_size,win_inc)

%nums : a vector of bins number that are desired to extract (for ex: nums= [4:7], return bins number 4,5,6,and 7)

% win_size : window size (for ex: 256 sample)

% win_inc : window increment (for ex: 128 sample)

[Ndata,f] = size(data);

epoch=floor((Ndata-win_size)/win_inc)+1;

n=9; % number of bins of bins

sw=1;

ew=win_size;

feat=zeros(epoch,f*n);

for k=1:epoch

Xext=hist(data(sw:ew,:),n) ; % histogram

% me(k,:)=mean(Xext); % mean

% va(k,:)=var(Xext); % variance

% sk(k,:)=skewness(Xext); %skewness

% ku(k,:)=kurtosis(Xext); % kutosis

feat(k,:)=reshape(Xext,1,f*n); %

sw=sw+win_inc;

ew=ew+win_inc;

end

s=0:11;

for i=1:n

h{i}=(feat(:,9*s+i));

end

out=cell2mat(h(:,nums));

% out=(diff(out));

% out=[va];