Envelope correlation between LFPs and Cortex: Negative values

Discussions
,
1

Hi all,

I am calculating the non orthogonalized envelope correlation of the Hilbert signal in the alpha range between the STN LFPs and the source-reconstructed cortical signal. I have read that you should not expect negative values, and this is the case when I do env corr between cortical patches or between sensors; yet, when I do it between LFPs and cortex, negative correlation values appear. How come? How should I interpret those?

Thanks for your continuous help,
Marco

2

Hi @marco1994,

This is true, only for the case when orthogonalization.

This is because the overall magnitude of the correlation between x and y is computed as the mean of the correlation of the envelop of x and the envelop of y orthogonalized with x (i.e., y_ort_x), and the correlation of y and x orthogonalized with y:

[abs(corr(env(x), env(y_ort_x))) + abs(corr(env(y), env(x_ort_y)))] / 2

This is done as corr(env(x), env(y_ort_x)) is not the same as corr(env(y), env(x_ort_y))

If not orthogonalization is used, the corr(env(x), env(y)) is equal to corr(env(y), env(x)), so the sign can be kept.

3

Thanks for your explanation!
Marco