Code for Inverse Problem Optimization

Discussions
1

Hello,
I am trying to understand the Computation behind the Inverse Problem

In this case, I am focusing on MNE since it seems to be the easiest.
From what I understood, first data is projected on the Head.Gain Matrix.
Then Haad.Orientation and Head.weight are computed which eventually provides weighted and whitened matrix of head consisting data.
Now at this point, I was expecting to see some sort of the loop that performs optimization of the famous cost function that MNE have, however, I saw the following line which basically uses the lead field matrix and its SVD to compute the kernel.
Can somebody help me and tell me where exactly the optimization of the cost function occurs in the code for the simple case of MNE with current density map?

Thank you
Best
Younes

image
image1239×642 145 KB

2

@John_Mosher, @Sylvain?

1 Like
3

Hi Younes:

There is no loop with MNE as the solution is unique and can be written analytically. Please refer to the technical writing in the following paper for details:
https://www.researchgate.net/publication/3321395_Electromagnetic_Brain_Mapping

Brainstorm implements one version of the weighted minimum-norm optimizer, with weights based on source depth and noise covariance, which can be changed from the Expert mode of the source imaging panel.

Hope this helps

1 Like