DRAFT, 10/03/01

W. Ross Adey, M.D.

Physiology Dept

Loma Linda University School of Medicine

White Paper:

SPECIAL FOCUS AREA: BRAIN MACHINE INTERFACES

(Defense Sciences Office, Solicitation BAA 01-42, Addendum 1, 9/17/2001)

1. Statement of Target Goals:

The broad goal of this DARPA plan is to create new technologies that would augment human performance through noninvasive access to brain signal patterns in real time, for the purpose of using these integrated brain signals in sophisticated control of certain machine operations:

1) Extraction of neural codes related to patterns of sensory and motor activity in a spectrum of motor activities ranging from simple to complex; and to develop similar capabilities with respect to brain signaling in the special senses, specifically in the auditory and visual modalities.

2) Direct feedback to the brain from peripheral devices or systems of derived or transformed signal patterns that would allow closed-loop control of robotic or other peripheral systems.

3) Develop and apply new pattern recognition techniques to physiological patterns of brain activity derived noninvasively, applying outputs of these analyses to close loop control of peripheral devices. Identification of these patterns would involve tests of multimodal data acquisition, including transduction by magnetic and light sensors.

4) By the use of new materials and new designs for devices, to meet needs in neural control of systems having defined physical elastic and compliance characteristics; and to develop working prototypes of these devices and systems.

5) Demonstration of adaptive abilities in the patterns of neural signals selected for device control, with progressive refinement of neural signal pattern sensitivity and selectivity, as a biological self-organizing property.

6) Robotic or similar implementation of an actual working system of controllers that incorporate neural sensory and motor control, based on force dynamic and sensory feedback.

This White Paper focuses primarily, but not exclusively, on TOPICS 1-3.

2. Extraction of neural codes related to patterns of sensory and motor activity

A non-invasive approach to detection of relevant neural signal patterns is offered as a preemptive guideline for this program. Thus, this will become the first and highest

-1-

priority in development of all signal conditioning and data analysis systems. It also determines the nature and scope of key higher order factors to be evaluated in the planned studies.

2.a. System requirements for development of neural codes:

2.a.1. It will be assumed that idiosyncratic signatures unique to an individual

subject will be unacceptable for machine control. Acceptable neural codes

should contain robust signatures. Optimally, this would permit their

adaptive utilization by a population of suitably trained individuals.

2.a.2 Acceptable neural codes should be exportable and capable of operation

world wide, either as local controllers, or as telecontrollers physically

removed from the operational environment.

2.a.3 Genesis of an approved neural code by a trained subject, and application of

its transforms to machine control, should be via devices functioning as far

as possible on a non-interference basis with essential physical mobility of

the subject. Thus, beyond proof-of-concept technology in a laboratory

environment, major attention must be devoted to design, development and

testing of operational hardware.

3. Evidence for existence of control signatures in electroencephalographic (EEG)

records of human task performance under controlled conditions

The human EEG is a wave-like process generated in the cerebral cortex under all conditions of wakefulness and sleep. The planned studies require a clear understanding of structural organization mediating its genesis. It is a phenomenon unique to cortical structures and to a few non-cortical central nervous centers with a similar anatomical infrastructure.

3.a.1 Cortical neurons have large, spreading dendritic branches, or

arborizations, which are entwined with similar arborizations of adjacent

neurons. They make physical and functional contact with one another.

3.a.2 Electrodes placed inside single cortical neurons record large (~20-30

millivolts) slow electrical waves with most energy in the spectrum

1-30 Hz. These waves are generated in the dendritic trees (Elul, 1972),

and have led to the eponym, "the independent dendrite" (Szuromi, 2001).

3.a.3 A small portion of this intracellular signal (< 1.0%) appears in brain fluid

that surrounds neurons. It is recorded as the summed activity of dendrites

of many neurons on the cortical surface, or on the scalp, with little further

attenuation (10-200 microvolts).

3.a.4 It is important to note that the EEG is not formed as the envelope of the

firing of numerous nerve impulses. It is an intimate signaling system by

which a population of nerve cell cells in a single region or domain

"whisper together" in a faint and private language (Adey, 1993).

3.a.5 Formerly dismissed as "the noise of the brain's motor" by the Nobel

-2-

Laureate Sir John Carew Eccles, the EEG is now recognized as intimately

correlated with information transaction and recall. Genesis of nerve

impulses may be considered a high order transform of this transactional

wave activity (Wei et al., 2001).

3.a.6 Moreover, 60% of cortical neurons are Golgi Type II cells. Lacking an

axon for nerve impulse transmission, they can communicate with adjacent

neurons solely through dendro-dendritic contacts.

3.a.7 There is growing evidence that such an internal communication system is

a determinant of a response threshold for a group or domain of elements

(Bialek, 1983, 1984) in quorum decision making, and is not the property

of any single element within that domain (Jessup et al., 2000).

3.b. As discussed below, by reason of its genesis in multidimensional patterns

that appear in scalp records as an almost infinite variety of spatiotemporal and phase-related patterns, the EEG would appear to offer a much richer repertoire of potential control signals than available from multineuronal firing records, even if these were available for the current purpose.

3.b.1 Early animal studies of correlates between intraneuronal waves and the

EEG in the surrounding brain tissue confirmed a close similarity in

frequency spectra, but virtually no coherence between the two waveforms 3.b.2 Also, multipolar EEG records using orthogonal electrodes with tips

separated only by cellular dimensions (~10 microns) were incoherent in

their frequency spectra (Elul, 1972).

These findings support the conclusion that in normal brain tissue, individual

neuronal wave generators are in a chaotic or "noisy" state, as indicated by

strong trends towards Gaussian EEG amplitude distributions in man in

"resting" behavioral states.

3.b.3 Further studies in man have reported 1) sensitive trends away from

Gaussian amplitude distributions during mental task performance; and 2) a

virtual absence of Gaussian distributions in records from epileptic brain

tissue. If Gaussian EEG amplitude distributions provide indices of

functional levels of interconnection within and between domains of brain

tissue, they would support a model of the brain as a "noisy processor",

with this pseudorandom organization as a sensitive indicator of

information processing (Adey, 1972).

4.0 Experimental evidence for individual and group EEG signatures correlated with task

performances.

Development of new knowledge in this field of EEG control has stagnated following an era of substantive early achievements three decades ago. Most of this early work was funded by DOD agencies - AFOSR, ONR, ARPA - and by NASA. In part, this approach

-3-

fell into disfavor through problems of neuroelectric data acquisition in performing subjects, in computational limitations in nonlinear analysis of chaotic systems, and limited robustness of pattern recognition methods.

Related developments in robotics and in medical prosthetics have come to rely heavily on neuromuscular control signals. These have a high level of reliability and repeatability. Thus, the potential for successful brain-machine interfaces has remained largely unexplored in the interim, and modern data analysis and pattern recognition methods have not been applied to these problems.

4.a. Shared EEG signatures in a population of astronaut candidates as concomitants of

auditory and visual tasks of increasing complexity

Complexities of multichannel EEG records and differences between subjects have discouraged fine interpretations based on visual inspection of computer printouts. With support from NASA, a "normative library" was therefore developed by computer analysis and pattern recognition, using EEG data from 50 astronaut candidates (Walter et al., 1967). Scalp EEG electrodes were placed according to a modified International 10-20 Pattern to provide 18 data channels.

Accurately timed physiological stimuli and perceptual and learning tasks were presented, thus establishing group means for EEG records from each test situation. In each case, despite wide individual differences between subjects, the group mean and/or pattern of variance in spectral densities for each test condition presented a characteristic pattern. These patterns were consistent with neurophysiological organization in corticosubcortical interrelations of cerebral systems.

There have been no published replications of this or similar studies, despite its clear potential for monitoring and controlling man-machine interactions in complex environments.

4.b. EEG motor signal tracking with an adaptive phase-locked loop as an approach to

prosthetic control

A study at UCLA Brain Research Institute, supported in part by AFOSR, established the feasibility of prosthetic control by the use of EEG signatures (Nirenberg et al., 1971).

The study involved two subjects, one normal and the other a hand amputee. As a prototype for control by an adaptive phase-locked loop (PLL), each subject performed a prototype task of opening and closing the hand. The normal subject accomplished this, while the amputee attempted to use the missing hand as though it were present. Simultaneously, the scalp EEG was recorded from a site presumed to overly the left motor cortex. Parameters of the PLL were adaptively chosen to implement on-line monitoring of the EEG. Threshold logic applied to the PLL output V yielded a promising method to detect motor action from the EEG record in the normal subject and the amputee. The PLL parameters, optimized by visual inspection of V, were identical in both cases, except for the center frequencies of the voltage-controlled oscillator. The final

-4-

thresholds differed only slightly.

The investigators remarked that at that time, "The control of externally powered prosthetic devices by sensing of brain waves has been completely ignored." The position remains unchanged three decades later.

**********************************************************************

[5. Method of procedure, expected outcomes, utility of the method…….etc., ….etc]

**********************************************************************

References

Adey WR. Organization of brain tissue: is the brain a noisy processor? Internat J.

Neurosci. 3:271-284, 1972.

Adey WR. Electromagnetics in biology and medicine. International Union of Radio

Sciences. In: H Matsumoto, ed. Modern Radio Science. Oxford University Press.

pp.231-249, 1993.

Bialek W. Macroquantum effects in biology: the evidence. Ph.D. thesis, Department of

Chemistry, University of California, Berkeley. 250pp. 1983.

Bialek W, Wit HP. Quantum limits to oscillator stability: theory and experiments on

acoustic emissions from the human ear. Phys. Lett. 104A:173-178, 1984.

Elul R. The genesis of the EEG. Internat. Rev.Neurobiol. 15: 228-251, 1972.

Jessup JM, Fra nz M, Sonmez-Alpan E, Locker J, Skena K, Waller H, Battle P, Nachman

A, Weber ME, Thomas DA, Curbeam RL, Baker TL, Goodwin TJ. Microgravity

culture reduces apoptosis and increases the differentiation of a human colorectal

carcinoma cell line. In Vitro Cell Dev. Biol. - Animal 36:367-373, 2000.

Nirenberg LM, Hanley J, Stear EB. A new approach to prosthetic control: EEG motor

signal tracking with an adaptively designed phase-locked loop. IEEE Trans.

Biomed Eng. BME-18:389-398, 1971.

Szuromi P. The independent dendrite. Science 293:2166, 2001.

Walter DO, Kado RT, Rhodes JM, Adey WR. Electroencephalographic baselines in

astronaut candidates estimated by computations and pattern recognition

techniques. Aerospace Med. 38:371-379, 1967.

Wei DS, Mei YA, Bagal A, Kao JPY, Thompson SM, Tang CM. Compartmentalized and

binary behavior of terminal dendrites in hippocampal pyramidal neurons. Science

293:2272-2275, 2001.