Surface Electromyography in Network Spinal Analysis
This research, which involves human subjects, has been approved by the
University Park Institutional Review Board (IRB)
of the University of Southern California
(case USC UPIRB #01-01-009)
Network Spinal Analysis (NSA) is a technique through which the practitioner applies light pressure at precise points along the spine closely correlated to regions where dural attachments to the bony vertebrae and contiguous structures have been demonstrated. Although patient specific, within one to three months of care, the spine exhibits a spontaneous rocking motion which is not initiated voluntarily, but can be voluntarily ceased. Retrospective and longitudinal studies have indicated that the benefits of this type of care include enhanced flexibility and physical stability of the spine. This type of care also appears to relief adverse mechanical tensions in the spinal cord. Adverse Mechanical Tensions in the Central Nervous System is a theory developed by Alf Breig, a neurosurgeon from Stockholm, Sweden. Most importantly, this kind of care seems to provide the same kind of repetitive motion that has allowed, in the well publicized case of Christopher Reeve, some late sensory and motor recovery from spinal cord injury.
The theoretical foundation of Network Spinal Analysis is Alf Breig's theory of Adverse Mechanical Tensions in the Central Nervous System. In this theory, the attachment of the dura mater to the cervical vertebra creates pathological tensions on the cord in case of vertebral misalignment or postural problems. In Network Spinal Analysis, the dural-vertebral attachment is put to use to creates an oscillation that produces the same kind of repetitive motion that has been linked to spinal cord injury recovery in the well publicized case of Christopher Reeve. The oscillation is first localized to the vertebral area but soon propagates down the spine. Likewise, the attachment of the filum terminale to the coccyx in the sacral area also creates an oscillation that propagates up the spine. The uprunning and downrunning waves along the spine eventually yield the rocking motion of the spine typical of NSA. In addition to its potential applications to spinal cord injury recovery, in less dramatic cases it provides an intensive exercise for the back musculature, not reproducible by classical physio-therapeutical means.
During the rocking motion, a fair amount of surface electromyographic (sEMG) activity is present along the spine. The sEMG signals recorded at the cervical, thoracic, lumbar, and sacral levels show "bursts of EMG activity" appearing at random and lasting anywhere from a few seconds to a minute.
Early sEMG nonlinear signal modeling
Our early attempt to achieve a better understanding of this neurophysiological phenomenon has been to perform a nonlinear dynamical modeling of the sEMG signal during its "bursting" phase. The False Near Neighbor (FNN) approach, popularized in the chaos literature by Ruelle as a mean of estimating the dimension of an attractor, did not manage to reveal anything substantially different from a pseudo-random sequence. However, such proven techniques as the Nonlinear Canonical Correlation Analysis (CCA) and the Alternating Conditional Expectation (ACE) did reveal the presence of nonlinear phenomena and both techniques consistently pointed to a nonlinear dynamics of a dimension somewhere around 6 or 7. The CCA reveals a subtle, but definite (i.e., jumping from 0.1 to 0.8), increase of the 3rd, 4th, and 5th canonical correlation coefficients under nonlinear distorsion of the past and the future, while the ACE reveals some saturation in the regression functions predicting the future from the past.
Switching ARIMA modeling
Despite the nonlinearities in the sEMG signal, the possibility of a linear modeling was comtemplated, first because it could clarify the issue of the stationarity of the signal, and also because with the marked "bursts on top of background" signal aspect, it appeared plausible that the signal would be fairly well represented by two linear models (one for the burst, the other for the background) switching between each other. The background versus bursting discrimination was done on the base of the correlation and partial correlation sequences. Surprisingly, the background signal appears less stationary, as its correlation sequence does not decay as fast as it should be. The burst signal, however, appears more stationary, as its correlation sequence decays faster. A switching criterion, based on the correlation, was developed, and it appeared that the overall signal is fairly well modeled by the two linear models swithing among each other.
Spatio-temporal analysis of wave phenomenon
Next, a "spatio-temporal" analysis of the collection of signals recorded at various points was developed to determine whether a "wave" pattern traveling along the spine is present.
The first method relies on the Canonical Correlation Analysis of the past of the raw signal at one point S (the "source") and a time-shifted version of the future of the signal at another point T (the "target"). More specifically, from the canonical correlation coefficients, the Akaike mutual information is computed versus the time-shift and, if the mutual information shows a maximum for a time-shift t_s, it can be concluded that it takes an amount of time t_s for the wave to travel from S to T. From this analysis, it appears that the "source" is the sacral area, for indeed, the correlation analysis shows an increasing amount of time for the wave to go to the lumbar, thoracic, and finally cervical area. The fact that the sacral area appears the "source" is probably related to the attachment of the filum terminale to the coccyx. Conversely, the correlation analysis did not reveal such a consistent pattern when the cervical area is viewed as the "source." This is probably related to the fact that the attachment mechanism is by far more complicated in the cervical than it is in the sacral area.
Next, another simplified analysis based on correlation between wavelet subband signals was developed. The reason for focusing on some specific subband signals is that it appeared that some of the subband signals are noise related and hence irrelevant. That part of the signal the most relevant to the wave phenomenon is the D_8 subband signal of the Daubechies wavelet decompositon of order 3 down to 8 levels. The correlation between the subband signals turned out to be much larger than the correlation between the raw signals, hence a higher confidence in our assertion that there exists a nonvanishing correlation between the signals at various points, hence a higer confidence in the assertion of existence of a wave phenomenon.
The spatio-temporal analysis was conducted on a quadriplegic subject who had sustained the same kind of neck injury as Christopher Reeve and who had been under NSA care for about a year and who had recovered some control of his fingers and toes. The purpose of this study was to confirm the spinal cord injury recovery by spatio-temporal analysis of the sEMG signals running along the spine. The same analysis was also done on a base-line subject, in the same experimental setup and under the same protocol as the quadriplegic subject. The quadriplegic subject shows a weaker correlation involving the neck signals, as can be expected, but existence of a correlation between such distant signals as the sacral and cervical signals, with a 99% confidence level, reveals some recovery from the spinal cord injury.
For more information about this project, contact Stephan Bohacek at firstname.lastname@example.org or Poonsuk ("Matt") Lohsoonthorn at email@example.com or Vikram Mahajan at firstname.lastname@example.org.
For Licenced practitioners only
1. S. Bohacek and E. Jonckheere, "Chaotic modeling in Network Spinal Anaysis: Nonlinear Canonical Correlation with Alternating Conditional Expectation (ACE): A preliminary report," in Journal of Vertebral Subluxation Research, vol. 2(4), pp. 188-195, Dec. 1998.
2. E. Jonckheere, S. Bohacek, and P. Lohsoonthorn, "Dynamic modeling of spinal EMG activity," NSF Southwest Regional Workshop on New Directions in Dynamical Systems, University of Southern California, Los Angeles, Nov. 16-19, 2000.
3. E. Jonckheere, P. Lohsoonthorn, and R. Boone, "Dynamic modeling of spinal electromyographic activity during various conditions," American Control Conference (ACC2003), Denver, Colorado, June 4-6, 2003, Session WA-13-3, Biomedical Applications, pp. 465-470. [power point presentation]
4. P. Lohsoonthorn and E. Jonckheere, "Nonlinear switching dynamics in surface electromyography of the spine," International Conference "Physics and Control" (Physcon2003), St. Petersbourg, Russia, August 21-23, 2003, pp. 277-282. [power point presentation]
5. E. A. Jonckheere and P. Lohsoonthorn, Spatio-temporal analysis of an electrophysiological wave phenomenon," International Symposium on the Mathematical Theory of Network and Systems (MTNS2004), Leuven, Belgium, July 5-9, 2004. [power point presentation]
6. E. A. Jonckheere, P. Lohsoonthorn, and V. Mahajan, "ChiroSensor: An array of noninvasive sEMG electrodes," Medicine Meets Virtual Reality (MMVR 2005), Long Beach, CA, January 26-29, 2005; appeared in Medicine Meets Virtual Reality-13, The Magical Next Becomes the Medical Now, IOS Press, Technology and Informatics, Volume 111, Edited by James D. Westwood, Amsterdam, the Netherlands, ISBN 1-58603-498-7, pp. 234-236. [poster presentation]
7. E. A. Jonckheere, P. Lohsoonthorn, V. Mahajan, S. Musuvathy, and M. Stefanovic, "On a standing wave Central Pattern Generator," Biomedical Signal Processing and Control, submitted, 2007.
8. A. Hiebert, E. Jonckheere, P. Lohsoonthorn, V. Mahjan, S. Musuvathy, and M. Stefanovic, "Visualization of a stationary CPG-revealing spinal wave," Medicine Meets Virtual Reality (MMVR 2006), Long Beach, California, January 24-27, 2006.; appeared in Medicine Meets Virtual Reality--14, Accelerating Change in Health Care: Next Medical Toolkit, IOS Press, Technology and Informatics, Volume 119, Edited by James Westwood, Amsterdam, the Netherlands, ISBN 1-58603-583-6, 2006, pp. 198-200. [poster presentation]