Dr. E. Tognoli,
25 September 2008
HBBL meeting
From below the surface
Inverse problem, raw EEG
and re-reference
Inverse problem: practical problem of head models
Smear the signal
Displace the signal
Inverse problem: practical problem of head models
Inverse problem: practical problem of head models
Working at the surface
(for now)
Why reference: sense the noise
• Principle of differential
amplification: the CMR
– (Signal + noise) – (noise)
• Take a scalp electrode (say C3) and
a fixed point (GND)
• Measure one potential difference
• Take a reference electrode (say
MA1) and a fixed point
• Measure a second potential
difference
• (Signal + noise) – (noise) = “a very
clean” signal
COMPARISON OF SPECTRAL ANALYSIS
ACROSS A RANGE OF REFERENCES
In most circumstances, topography, frequency and amplitude
are preserved across different references/re-references
However…
1. Peaks absent in linked mastoids
Peaks absent in linked mastoids
Peaks absent in linked mastoids
Peaks absent in linked mastoids
Peaks absent in linked mastoids
Peaks absent in linked mastoids
*
2. Peaks absent in average reference
Peaks absent in average reference
Peaks absent in average reference
3. Translocated peaks
Translocated peaks
Translocated peaks
Translocated peaks
4. Spectral shift
Spectral shift
5. Relative amplitude
(relative) amplitude
6. Degenerated re-references
Degenerated re-references
Bad average reference
Bad mastoids reference
Conclusion
Mastoid (especially re-reference)
Masking: alters temporal (peripheral) components.
(Note however that EGI net does not support
mastoid attenuation: electrodes are not exactly
positioned over the mastoid bone)
CZ or average reference
Masking: alter vertex components.
Splitting: tendency for competition of power: two peaks cannot dominate at a single
frequency.
Sparseness: The amplitude distribution of peaks is sparser on top, accordingly more
difficult to separate from noise when amplitude is small.
Of peaks and troughs (~): average reference is much better than other techniques to
see power depression.
Overall: However, in this study, less frequent missed peaks in average reference than in
linked mastoid re-reference.
Who is right?
Some of the peaks from one montage may be spurious, always check supra- and subharmonics, organization, and consistency across conditions and temporal windows.
Initial positioning of the reference
No single good recipe, depends on the study. Yet it is paramount to get the initial
reference right.
Anatomical wisdom: choose your reference electrode in an intelligent fashion with
respect to your particular studies, and the anatomical structures it is expected to
recruit. For instance, it does not make a lot of sense to study the motor cortex and
record with a Cz ref or linked mastoids if you expect temporal (e.g. auditory) activity.
Empiricism: In pilots, try two or three references.
Symmetry: if you want to study hemispheric lateralization, you will need suitable
bilateral or medial reference .
Impedance matching: if you choose a pair of electrodes as reference, it is important to
get them at a low and matched impedance. Ideally, the EEG amplifier should offer
digitally linked reference leads.
Re-reference
Most components will be robust to re-referencing, exhibiting minimal spatial shift
and minimal amplitude, frequency… alteration.
The examples I presented today were extreme cases that occur less frequently,
but a serious investigator needs to keep an eye on those facetious spatial biases
of the raw EEG.
In the exploratory phase of your analysis, always try at least one complementary
re-referencing to unmask hidden components from the vicinity of the reference
electrode.