i wanted the slope with respect to time frame

I wanted to calculate slope of channel 1 to 15 with respect to the time frame. The values in the tables are HbO values which should be Y axis and X axis should be time time frame which in this case is 1510.

I have attached my data file as well.

How do i calculate the slope of channels 1 to 15 indivijually and place the values in a different table and perhaps even plot to visually see it????

• Slope: the value that fits a regression line to the given data set.
• https://journals.plos.org/plosone/article/figure/image?size=medium&id=info:doi/10.1371/journal.pone.0208843.g007

Something similar to the url i have posted above.

Thank you

[Jan]

The question is not clear. What are "channels 1 to 15"? Why did you highlight the first cell? What are "HbO" values? Where do we find the "time frame"?

[CalebJones]

Jan HbO is Heomoglobin values. Highlighted 1st column is HbO amplitude for channel 1 which is outputed from the machine.

Time frame is row index which starts at 1 and ends at 1510.

So 15 columns shows HbO amplitude of 15 channels.

So i wanted to perform polyfit func on curve from channel 1.

So one by one i wanted to calculate the slope of each channel so.

Answer 1

[Star Strider]

First, negative values for haemoglobin or oxyhaemoglobin do not make sense physiologically.

I have no idea what you want to do, so start with:

D = load('HbO_Good_channels.mat');
HbO = D.HbO_good_channel;
Ts = 35/size(HbO,1);                                                    % Create A Sampling Interval, Since None Are Provided
T = linspace(0, size(HbO,1), size(HbO,1))*Ts;                           % Time Vector
lgdc = sprintfc('Ch %2d', 1:size(HbO,2));                               % Legend String Cell Array (Channels)

figure
plot(T, HbO)
grid
xlabel('Time')
ylabel('HbO')
legend(lgdc, 'Location','eastoutside')

for k = 1:size(HbO,2)
    cfs(k,:) = polyfit(T(:), HbO(:,k), 3);                              % Coefficient Vectors: ‘polyfit’
end

figure
hold all
for k = 1:size(HbO,2)
    pf(:,k) = polyval(cfs(k,:), T(:));                                  % Evaluate Fitted Polynomials
    plot(T, pf(:,k))
end
hold off
grid
xlabel('Time')
ylabel('Regression Fit')
legend(lgdc)

Experiment to get the resultl you want.

[CalebJones]

Your a Genius Mate... Star Strider This is what I'm getting.

All this is fNRIS signal.

I took the sampling rate as 25 since that was interval of activity but again I'm confused.

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0208843 this paper mentions Slope: the value that fits a regression line to the given data set.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5788966/ The slope was computed using the least-squared linear regression with the polyfit matlab function.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5509792/ fNIRS features were based on HbO and HbR amplitude (HbO/R Amp.), slope of HbO and HbR (HbO/R slope).

None of papers mention how they calculated and I'm implementing all this by myself.

Basically I am doing this to find best feature which i can use to train my model..i have finished mean, struggling with slope at the moment.

[Star Strider]

Thank you!

I scanned the articles, however I didn’t see anything specific about the slope calculations. If you want the derivative of the polyfit results, use the polyder function to give you the slope as a function of the independent variable. (Use the polyder result with polyval, as you would the polyfit results.) See if that does what you want.

Please point me to specifically-relevant parts of those articles, in case I missed something. (Also, the one reference I saw used MATLAB R2015b. Several significant additions and improvements have been addes since that could make your analysis easier.)

[CalebJones]

Star Strider thanks a lot for your help.... ummm polyder can be used in place of polyfit??

I mean just replace polyfit by polyder?

And another question Is there a GUI or something that would calculate mean,kurtosis,slope,Std Devi,etc?

Imagine doing this for 20+ datasets...it would drive me nuts

i just have to show which feature would give me highest accuracy.

[Star Strider]

Use polyder on the output of polyfit, the polynomial created by polyfit. Then use polyval to create an evaluated vector to plot the derivative of the polynomial created by polyfit, the same way as you would use polyval on the original polyfit output, if you need to evaluate it and plot it.

The slope would be given by the polyder polynomial, evaluated at the appropriate points with respect to your independent variable.

If you have the Statistics and Machine Learning Toolbox, use the kurtosis function to evaluate your vector. See the documentation for details. The other functions are mean and std.

Taking one channel to demonstrate the derivative:

Ch1 = HbO(:,1);                                                         % Example Channel
P = polyfit(T(:), Ch1, 3);                                              % Polynomial Fit
Pf = polyval(P,T);                                                      % Evaluate Polynomial
dP = polyder(P);                                                        % Derivative Polynomial
dPf = polyval(dP,T);                                                    % Evaluate Derivative Polynomial

figure
plot(T, Pf,    T, dPf)
grid
legend('Polynomial Fit', 'Derivative of Polynomial Fit', 'Location','SE')

That should get you started. Experiment to get the result you want.

[CalebJones]

Star Strider This is what I am getting. So i should take dP or dPf as feature? So confusing.....

[Star Strider]

I don’t know.

I am providing you with the ability to calculate slope and other statistics. You have to decide on the features.

[CalebJones]

Okay Cool Star Strider Thanks for help so far....your a life saver dude...

Okay so if i wanted to calculate area under the curve T(he Area Under the Curve (AUC) was calculated by summing the absolute values of the signal. AUC = ∑ |x|)

Where X is again my HbO_Good_channel.

I'd have to use

Ts = 25/size(HbO,1);

T = linspace(0, size(HbO,1), size(HbO,1))*Ts;

T_transpose = T.';

AUC = area(HbO_good_channel(:,1),'LineStyle',':');So the values under AUC can be used as feature right?

[CalebJones]

Star Strider hey man this is AUC of all 15 channels and when i do trapz 1 x 15 values.....So am i supposed to use this as a feature ?

For better understanding https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5788966/figure/F4/ .

[Star Strider]

‘So am i supposed to use this as a feature ?’

You have to decide that. I just skimmed the articles to see if I could understand the MATLAB references so I could help you with the calculations.

The trapz function is the correct way to calculate the areas between the zero baseline reference and the values of the data.

[CalebJones]

You have been really helpful man... appreciate it dude....

Star Strider hey could you provide me your personal email address ,if you don't mind? I tripping over many hurdles and you're there for my rescue....if that makes sense.

[Star Strider]

Thank you. My pleasure.

I prefer to keep everything regarding MATLAB Answers posts here on the forum. Otherwise, there are discontinuities that create problems for anyone who later reads the threads. This is a general resource, not restricted to specific contributors.

[CalebJones]

Star Strider Sounds great!

[CalebJones]

Star Strider

There are markers pointing out the actvities rest and active.

~25sec is the window for rest and 25 for active and sampling freq Fs = 7.825 so each window is 7.825 x 25 = ~200 frames

Rest then active

Rest then active

Rest then active

3 Rest and 3 active

So above Ts = 35/size(HbO,1); instead of 35 it should be 25 right?? And while ploting i wanted to show on the x axis as rest and active states(like the picture below)

[Star Strider]

I have no idea.

[CalebJones]

If you have a look at this image. Like wise even i have extracted HbO which is oxy hemoglobin and HbR which is deoxy Hemoglobin.

And in my dataset is a task like rest active above image shows rest and tap.

Above image plots only 1 channel signal , likewise there at 15 channels.

My dataset(attached above) has calculated HbO Values and labeled things as rest and active for each channel.

So now i wanted to train my model such that it predicts rest and active states with decent accuracy using features like slope,mean,kurtosis,skewness,etc.

https://sci-hub.tw/https://www.ncbi.nlm.nih.gov/pubmed/27252637

https://sci-hub.tw/https://ieeexplore.ieee.org/document/6181455

Star Strider have any idea how to go about this ?

[Star Strider]

I do not have access to the IEEE article. If you want to upload a .pdf of it, I will look at it.

The Frontiers in Human Neuroscience paper uses linear discriminiant analysis (LDA) for its classifications. Use the classify function. There are a number of related functions linked to in and at the end of that documentation, so explore them as well. It might also be beneficial to use the pca function to understand the important aspects of your data. The paper otherwise appears to have everything you need to pre-process your data.

[CalebJones]

Star Strider ,

https://www.frontiersin.org/articles/10.3389/fnhum.2016.00237/full

One is a PDF and other is a link. I don't know the stratergy behind the training and accuracy prediction part.

[Star Strider]

Thank you for the paper.

I have no recent experience with the details of classification, since I have not used it in a while, and SVM was not yet available (at least to my knowledge) the last time I did extensive classification.

Rather than using ‘libSVM’, use the procedures outlined in Support Vector Machine Classification. (These first appeared in R2014a, so were not available when that research was done.)

[CalebJones]

Star Strider ,

Is this right way ?

I have a mat file below of the dataset.

You have to predict rest or active!

[Star Strider]

I have no idea. As I mentioned before, I have very little recent experience with classification, and essentially no experience with SVM.

I suggest that you open a new Question on this.

Answer 2

[Jan]

Maybe all you need is to call the gradient(X.') function, where X is the complete matrix?