View
222
Download
2
Tags:
Embed Size (px)
Citation preview
Spatial patterns of EEG gamma and fMRI
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Using the neurodynamics of local mean fields (EEG) manifested in electromagnetic potentials to interpret spatial patterns of cerebral blood flow in
behavior.
Walter J FreemanDepartment of Molecular & Cell Biology
University of California at Berkeleyhttp://sulcus.berkeley.edu
Organization for Human Brain MappingSendai, Japan 5 June 2002
Abstract: Dendrites vs. Axons: Energy requirements
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Dendrites take 95% of the energy that brains use to process information, whereas axons take only 5%. Their activity
is the main determinant of patterns in fMRI.
Dendrites are also the main source of the electric current that generates the EEG in passing across brain tissue.
The EEG has optimal temporal and spatial resolution for imaging neural activity in cognition, in order to relate it to
patterns of metabolic activity by means of fMRI.
Questions to be raised
OHBM’02.Sendai EEG and fMRI Walter J Freeman
1. What is the dependence of metabolic energy usage on the temporal spectral ranges of the EEG?
1 - 7 Hz — delta, theta?8 - 25 Hz — alpha, beta?25 — 100 Hz - gamma, higher?
2. What spatial structures of the EEG are best correlated with patterns of metabolic energy utilization?
Localization of specific psychological functions? Global patterns of cognitive operations?
Rabbit EEG, Temporal PSD
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Temporal Power Spectral Density, PSDt:
• Spectral peaks of power indicate limit cycle attractors, that are characteristic of band pass filters operating at single frequencies.
• Spectral distributions of power indicate chaotic attractors, that indicate nonconvergent, creative neurodynamics.
• The more revealing spectral displays are done in log-log coordinates: log power versus log frequency.
A. EEGs from olfactory bulb and visual cortex of rabbit superimposed on respiratory cycles.
B. Spectra show “1/f” fall in log-log coordinates, but with peaks in theta and gamma ranges for bulb but not so clearly for neocortex.
Spectral peaks vs. “1/f”
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Rabbit EEG, Spatial PSDx
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Left: Olfactory bulb. The upper curve is the spectrum of a point dipole. The dots show the spectrum of the 8x8 array. Right: Visual cortex. The Nyquist frequency is estimated to be 0.5 cycles/mm; sampling rate should exceed 1/mm.
EEG from the superior temporal gyrus recorded with a 1x64 linear array of electrodes spaced at 0.5 mm and 3.2 mm in length, fitting onto the gyrus without crossing sulci. These 15 adjacent EEGs are representative of the set. Note the fine-grain spatial differences.
Human intracranial EEG under anesthesia
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Another patient was recorded under local anesthesia, showing the emergence of gamma oscillations. The 1x64 linear array was held on the pia of the precentral gyrus for several seconds of recording.
Awake intracranial EEG
OHBM’02.Sendai EEG and fMRI Walter J Freeman
PSDs are compared from the EEGs in anesthetized and awake neurosurgical patients. Both reveal “1/f”.
Human temporal PSD
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Human pial spatial spectrum
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Spatial spectra of the human epipial EEG. These curves provide the basis for fixing the spatial sampling interval.
patient intercept, a slope, b max, c min, d inflection, x inflection, y c - d
1 4.13 -1.54 3.50 1.74 0.040 0.56 1.76
2 5.12 -2.27 4.24 1.90 0.038 0.41 2.34
3 4.42 -1.88 3.61 1.82 0.042 0.38 1.79
4 4.38 -2.20 3.70 1.48 0.032 0.33 2.22
5 4.16 -1.95 3.32 1.43 0.042 0.39 1.89
Average 4.44 -1.97 3.67 1.67 0.039 0.41 2.00
± SD 0.18 0.14 0.15 0.09 0.002 0.03 0.12
Table 1. Evaluation of spatial spectra by linear regression using 3 line segments.
log p = c f< x c/mm,
log p = a + b log f x ≤ f ≤ y c/mm,
log p = d f > y c/mm,
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Calculation of spatial Nyquist
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Design of an optimized epipial intracranial array
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Breakdown by temporal band, 5 Hz bands
Human scalp EEG and EMG
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Log-log display, EEG and EMG, temporal PSDt
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Log Frequency, Hz
LOG
POWER
Green: EEG from frontal scalp.
Red: Frontal EMG from scalp
Blue: EEG from parietal scalp
Black: EMG from parietal scalp
Human scalp EEG, EMG, Spatial PSDx
OHBM’02.Sendai EEG and fMRI Walter J Freeman
An example is shown of the human spatial spectrum from the frontal area of the scalp, with and without deliberate EMG.
Temporal band breakdown, 10 Hz bands
OHBM’02.Sendai EEG and fMRI Walter J Freeman
No significant dependence was found of the spatial spectra on temporal band width, except that for theta vs. gamma.
Spatial band breakdown, cycles/mm
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Temporal spectra are shown for narrow spatial pass bands in search for significant wave numbers. None were seen.
Derivative of PSD
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Left: Power as a function of frequency in linear coordinates. Right upper curve: log-log coordinates. Right lower curve: log of the derivative of the power vs. frequency, which may approximate the energy required for generating gamma.
The Answer to Question One
OHBM’02.Sendai EEG and fMRI Walter J Freeman
1. What is the dependence of metabolic energy usage on the temporal spectral ranges of the EEG?
1 - 7 Hz — delta, theta?8 - 25 Hz — alpha, beta?25 — 100 Hz - gamma, higher?
The answer is unknown. Insufficient data.
Studies are needed in which the fMRI patterns are carefully correlated with scalp recordings while power in the spectral
bands of the EEG and EMG is enhanced or diminished.
Spatial patterns of gamma EEG
OHBM’02.Sendai EEG and fMRI Walter J Freeman
A common approach to derive the behavioral correlates of gamma activity is to localize the ‘hot spots’ with high amplitude, fit them with an equivalent dipole, and find the phase relations between spots to infer causal relations.
An alternative approach is to combine both the high and the low amplitudes into a spatial pattern that resembles an interference pattern in fluids, and to follow sequences of these global patterns like frames in a movie film.
In patterns, dark spots are equal in value to light spots.
Spatial pattern measurement
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Measurement of spatial patterns of gamma activity is made easy by the fact that neuron populations in local areas such as sensory cortices by cooperative synaptic interaction form wave packets (Freeman, 1975), that share a common wave form in domains 10 - 30 mm in diameter (Freeman, 2002).
The textures of the wave packets are given by amplitude modulation (AM) of the gamma carrier wave. The phase modulation (PM) is useful to measure the size, duration, and location of wave packets, but it carries no information that relates to perception and cognition, and can be neglected in initial cognitive studies of global gamma activity.
Spatial AM patterns in the olfactory bulb
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Clustering of AM patterns with CS+ vs. CS-
OHBM’02.Sendai EEG and fMRI Walter J Freeman
A rabbit was trained to discriminate two odors from the background air (control, •), one that was reinforced (+), the other not (-), Each symbol shows a single pattern of AM modulation of the gamma carrier, which was projected from 64-space by stepwise discriminant analysis into 2-space.
Spatial filter tuning
OHBM’02.Sendai EEG and fMRI Walter J Freeman
The classification assay was used to find the optimal values for low and high pass spatial filters. The high cut-off was at the upper inflection in the spatial spectra. The low cut-off was fixed by the array window size.
Spatially distributed CS information
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Deletion of a subset of channels that was selected randomly degraded the goodness of classification. Information is uniformly distributed.
Spatial patterns, vision
OHBM’02.Sendai EEG and fMRI Walter J Freeman
A single set of 64 EEG traces; amplitude is in
upper right frame.
Classification of AM patterns: CS+ versus CS-
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Effect of channel deletion on visual CS classification
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Deletion of subsets of channels that were selected randomly degraded the goodness of classification. Information is uniformly distributed.
EEGs simultaneously from limbic and sensory cortices
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Determining gamma range
OHBM’02.Sendai EEG and fMRI Walter J Freeman
The classification assay was used to find the optimal values for low and high pass temporal filters. The high cut-off was at the inflection to the noise plateau in the spectra. The low cut-off was determined by the intrinsic nonlinear cortical dynamics. Gamma was species-specific: cat: 35-60 Hz; rabbit 20-80 Hz.
Multiple cortices - deletions of selected areas
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Auditory cortex: Ohl, Scheich & Freeman (2001)
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Intracranial gamma activity
Four sets of data from epidural electrode arrays have shown that the spatial AM patterns of gamma oscillations relating to perception contain information that is spatially distributed and graded, and not localizable to point sources.
• Olfactory bulb in rabbit• Sensory neocortices in rabbit• Auditory cortex of Mongolian gerbil• Multiple sensory and limbic cortices of cat
Analysis of scalp EEG by others* indicates they are global.
Scalp gamma in perception is likely to be non-localizable.
OHBM’02.Sendai EEG and fMRI Walter J Freeman
OHBM’02.Sendai EEG and fMRI Walter J Freeman
2. What spatial structures of the EEG are best correlated with patterns of metabolic energy utilization?
Localization of specific psychological functions? Global patterns of cognitive operations?
The answer is unknown. Insufficient data.
Studies are needed in which the fMRI patterns are carefully correlated with gamma EEG patterns by means of multivariate
statistics in very high dimensional state spaces, while the cognitive contents are manipulated by psychophysical designs.
The Answer to Question 2
OHBM’02.Sendai EEG and fMRI Walter J Freeman
Acknowledgments
I am grateful for contributions from my students and postdocs over 45 years of research on intracranial EEG in animals and
humans, and now on scalp EEG from normal volunteers.
Their names are listed in my books and in our numerous publications that we have co-authored in refereed journals.
Support for this research has come from the National Institute of Mental Health, grant MH 06686, and the National Aeronautics and Space Agency, grant NCC 2-1244.