How to do segmentation on ECG signal for feature extraction

I have a 1-minute ECG signal. How do I segment it into 10s? The Fs is 250Hz

thank you in advanced!

Answer 1

[Star Strider]

Use the Signal Processing Toolbox buffer function.

Calculating ‘n’ is straightforward:

n = 10 * 250

producing:

n =
        2500

So 10 seconds of a signal sampled at 250 Hz is 2500 samples.

[Mohamad Aswad Bachtear Effendy]

i tried the code but how I can plot it?

[Star Strider]

You can plot it any way you want to.

One possibility:

D = load('cu01m.mat');
EKG = D.val;
Fs = 250;
t = linspace(0, numel(EKG), numel(EKG))/Fs;
n = 10 * 250;
bufEKG = buffer(EKG,n);
ofstEKG = bsxfun(@plus, bufEKG, (0:5)*max(EKG));

figure
plot((0:size(bufEKG,1)-1)/Fs, ofstEKG)
grid

producing:

It asppears to be a normal EKG. I see no obvious pathology.

.

[Mohamad Aswad Bachtear Effendy]

i want to do feature extraction so do i need to plot the frames individually? and how to calculate r-r intervals on the segmented signals?

[Mohamad Aswad Bachtear Effendy]

and how to do qrs complex on signals?

[Star Strider]

It is straightforward to plot the frames individually. It would likely be easiest to use subplot for this, however that is not the only option.

Use either findpeaks (with appropriate name-value pair arguments) or islocalmax (with appropriate name-value pair arguments) to detect the R-deflections, then do the necessary calculations to get the R-R intervals, and instantaneous rate. There may be other options as well, however those are the functions I generally use for these sorts of problems.

[Mohamad Aswad Bachtear Effendy]

is this the correct way to subplot it? I think it's wrong

subplot(2,2,1)
ofstECG0 = bsxfun(@plus, bufECG, (0:0)*max(filtered_ECG));
plot((0:size(bufECG,1)-1)/Fs, ofstECG0)

[Star Strider]

It is wrong!

Please think carefullly about what you want to do. Also, review the documentation for subplot (that I linked to).

Try this:

figure
bufSz = size(bufEKG,2);
for k = 1:bufSz
subplot(bufSz,1,k)
plot((0:size(bufECG,1)-1)/Fs, filtered_ECG(:,k))
grid
end

.

[Mohamad Aswad Bachtear Effendy]

there is an error saying the vectors must be in same length

[Mohamad Aswad Bachtear Effendy]

solved the error

the code is to detect R peaks. Is this correct?

[max,rr] = islocalmax(ofstECG, 'MinProminence',400);      
v = find(max);
RR = rr(v);

[Star Strider]

Not entirely.

Note that max is an important built-in function, and naming a variable ‘max’ overshadows it, so that you will not be able to use the function later in your code. Other than that (that you must correct), ‘v’ is correct. I wouild call it ‘Rpeak’ or something appropriately descriptive.

I have no idea what ‘RR’ does, since ‘rr’ returns the prominence values of the peaks. It has nothing to do with the R-R intervals. I would call it ‘Rprom’.

Other than with respect to max, the others are suggestions to make your code more descriptive and easier to understand later.

[Mohamad Aswad Bachtear Effendy]

is the ofstECG in the islocalmax is correct? because it seems that the R peaks are not detected in the frames

[Star Strider]

Note that ‘ofstEKG’ is simply designed to present all the records in one axes in the plot. You probably want to use the columns of ‘bufEKG’ instead, with each column processed individually. That would be easier. and would give the correct result for each segment.

[Mohamad Aswad Bachtear Effendy]

The amplitude ranges are different in each frames, how to adjust it so the amplitude range are the same ?

[Star Strider]

You are plotting ‘ofstEKG’. You should be plotting the columns of ‘bufEKG’. That will solve the problem.

[Mohamad Aswad Bachtear Effendy]

Sorry but why the QRS points are not in the correct places?

[Star Strider]

You need to analyse the columns of ‘bufEKG’ separately.

[Mohamad Aswad Bachtear Effendy]

[QSpeak,qs] = islocalmin(bufECG, 'MinProminence',0.1);      % Detect: Q & S
w = find(QSpeak);
QSprom = qs(w);
[Rpeak,rr] = islocalmax(bufECG, 'MinProminence',1.5);      % Detect: R
v = find(Rpeak);
Rprom = rr(v);

figure
bufSz = size(bufECG,2);
for k = 1:bufSz
subplot(bufSz,1,k)
plot((0:size(bufECG,1)-1)/Fs, bufECG(:,k))
hold on 
plot(t(v), bufECG(v), 'r^')
plot(t(w), bufECG(w), 'gv')
hold off
grid

legend('ECG', 'R', 'Q S')
xlim([0 10])
title('QRS marking')
xlabel('Time(s)')
ylabel('Amplitude(mV)')

end

I'm a bit blank. Which part need to be adjusted in order to analyse the frames one by one?

[Star Strider]

Use a loop. The islocalmin and islocalmax calls need to be in a loop so that ‘bufEKG’ is referred to by column:

for k = 1:size(bufEKG,2)
bufEKG(:,k)
end

I leave the details to you.

[Mohamad Aswad Bachtear Effendy]

The Q and S seems right but I cannot do the R peak.

[Star Strider]

Plot the peak values as a functioin of the time locations.

[Mohamad Aswad Bachtear Effendy]

Sorry but I'm don't quite understand about this

[Star Strider]

The findpeaks function returns [pks,locs] if you only call it with the individual columns of ‘bufEKG’.

The plot call for them for each column should then be:

plot(t(locs), pks, '^r')

That is all there is to it.

[Mohamad Aswad Bachtear Effendy]

Sir, how to find QRS duration? Do I need to replace islocalmin with findpeaks too?

[QSpeak,qs] = islocalmin(bufECG(:,k), 'MinProminence',0.05);      % Detect: Q & S
w = find(QSpeak);
QSProm = qs(w);

[pks,locs_R] = findpeaks(bufECG(:,k),'MinPeakHeight',0.5,'MinPeakDistance',1); %R peaks

[Star Strider]

The islocalmin function will provide the locations of the maximum negative extent of the Q and S deflections. I doubt that findpeaks will improve on islocalmin in that calculation, however it may be worthwhile to do that experiment.

However the QRS complex is defined as beginning with the departure from isoelectric reference immediately preceding the Q-deflection, and the return to isoelectric immediately following the S-deflection in a normal EKG, and the return to the level of the ST segment if there are ST-T abnormalities, as can occur with ischaemia or subendocardial myocardial infarction (ST-T depression), STEMI (ST-T elevation myocardial infarction), or drug or electrolyte abnormailites. So it is necessary to determine the point at which the Q-deflection begins and the S-deflection ends, in order to calculate those intervals correctly.

It is possible to use the maximmum negative deflection of the Q and S deflections to estimate the QRS duration, as determined by islocalmin, however that must be noted, since it is at variance with the standard definition of the QRS duration calculation, and will provide a shorter value for the QRS duration.

[Mohamad Aswad Bachtear Effendy]

Thank you very much for your guide sir.

[Star Strider]

As always, my pleasure!