What is LORETA neurofeedback?

To explain what LORETA neurofeedback is we have to start by looking at what EEG signals actually represent - and what they don't.

Despite EEG's long history, for most of that time we haven't really been clear about how signals detected at the scalp relate to actual local brain activity.  If only we could do this it would be important as it would allow non-invasive, inexpensive, insight into realtime brain activity.

EEG is a useful indicator of some aspects of brain function but it's not a direct measure of how information is being processed.

It's a bit like putting your hands on your computer, feeling heat, realising that processing is going on inside but not being clear about where exactly that processing is taking place.  LORETA is an attempt to link surface EEG signals to the brain areas that actually produced them. This is not a trivial problem as we shall see.

 LORETA relates surface EEG data to activity in specific brain regions

LORETA relates surface EEG data to activity in specific brain regions

What is LORETA?

LORETA is an acronym for low-resolution electromagnetic tomographic analysis. A bit of a mouthful for a technique first described in 2002 that attempts to link whole-cap EEG activity detected at the surface of the scalp with the area of the brain's cortex that gave rise to that activity. 

To understand how this is a challenge we need to describe what are known as the forward problem and the inverse problem. 

The forward problem in brain physiology suggests that given that we might know the sources of electrical potential and the brain's anatomical structures, it is possible to predict the external scalp potentials. This starts with the specific source and links it to the effect at the surface (a forward problem).  For one set of sources and anatomy the prblem has a deterministic solution. The solution shows that we understand how electrical potentials from neurons can produce the electrical signals we see at the scalp. We have known how to do this for a long time but starting with electrical potentials inside the brain means an invasive procedure.

The inverse problem of trying to deduce the region of the brain responsible for electrical signals measured at the scalp is not deterministic. In other words, there would be many (infinite) potential solutions and the problem becomes impossible to solve without making certain assumptions. Whilst assumptions were made in devising LORETA these assumptions have been subjected to independent validation.  Caution is always needed though.

LORETA - feedback based on brain region activity

The introduction of LORETA techniques to neurofeedback introduces new opportunities and also some demands.  The potential that arises is that we could provide feedback that relates to an actual brain region rather than being based on the potential measured at an electrode site. The feedback would be much more directly related to specific brain activity and potentially more effective.

A LORETA approach starts with a full cap EEG capture such as with the NeXus 32

In order to achieve meaningful data it is generally necesary to use all 19 sites in a standard 10-20 electrode cap.  The data must be artefact free to provide an uncontaminated solution.  LORETA computes a solution fo thousands of spatial regions referred to as voxels which make up the 3D volume of the brain. LORETA uses approximately 2,300 7mm voxels.

The LORETA solution is a matrix operation that converts the scalp data into the internal representation in each sample instant and this takes quite a bit of computing power. 

It's been estimated that at a data sample rate of 256 Hz and for 10 frequencybands, this amounts to over 90 million floating point computatons per second.

This computational load prevented real-time solutions and practical application in neurofeedback until relatively recently.  By using advanced game-oriented computing techniques it has become possible to provide the processing speed necessary to provide real-time feedback. LORETA based techniques can be combined with Z Score approaches to get an assessment and training of voxels based on normative data.