Abstract

"Towards non-invasive laminar electrophysiology"

The MEG data we measure are due to an interaction between cortical structure and function.  Typically we take structure (grey matter) to be known and attempt to reconstruct function (current flow). This problem is ill-posed and so different functional assumptions about how the brain might work (sparse or distributed, correlated or uncorrelated sources) give rise to different functional estimates. The problem is that we almost never know the functional ground truth. Typically therefore we use metrics like model evidence- which balances the amount of data explained by a functional estimate with its complexity- to judge between solutions. 

An alternative is to turn the problem around and pretend that the functional assumptions are known and attempt rather to reconstruct the anatomy. The advantage here is that we know the anatomical ground truth (where the grey matter lies).  If our functional assumptions are groundless we will have no chance of recovering the grey matter structure; if however our functional assumptions have some validity then our anatomical estimates should match the true brain structure.

There are many applications of this approach.  I will begin by talking about how we can use it to factor out errors in co-registration. I will then go on to get a measure of functional accuracy by seeing how much one has to distort a brain before it becomes an unlikely model.  Finally I will talk about how we can use these models of structure to distinguish between functional contributions from deep versus superficial cortical laminae. 

Read more about Professor Gareth Barnes:
http://www.fil.ion.ucl.ac.uk/Research/grbarnes.html

ALL ARE WELCOME

After the talk DNC offers coffee and cake in front of the auditorium.