principle component analysis for 4D images

Dear All, I am trying to apply PCA on 4D image, where the 4th D is time. and once I center the data (by subtracting from mean), I should get negative values, which is not acceptable in my case. First, how to get red of negative values. Second thing, I wish to get time back. Third, how to apply rotation to get the factors.

Any suggestions on these two things. Thank you so much.

Answer 1

[Image Analyst]

If you cast your images to uint8 or uint16 before subtracting, you won't get any negative values because they will be clipped off. (Not sure why you want to do this though).

Not sure what you mean by the "second and third things"

Answer 2

[mas saud]

thank you for your quick response. the data already in unit16, and as in PCA, one need to center the data, therefore, there should be a negative and positive around zero (center). I am dealing with activity (PET image), so I should not accept negative values. after centering the data, I name the resulted matrix imgc, then

imgv=cov(imgc);
[eigverctors,eigvalues]=eig(imgv);
eigverctors=eigverctors(:,end:-1:1);
imgr=imgc*eigverctors;

resulted image (imgr) has a components, if I consider first 2 components to be of interst, and others as a noise. the issue is, if I want to get for instance time activity curve from first component, I do not have the fourth dimension.