Shape Analysis of Somatosensory Evoked Potentials to Detect a Focal Spinal Cord Lesion

G. Agrawall , D.L. Sherman2, P. Wa1czak\ J.W.M. Bultel , N.V. Thakor\ D.A. Kerr\ A.H. Ani IJohns Hopkins University, Baltimore, MD 21205

2Infmite Biomedical Technologies, Baltimore, MD 21211

Abstract-In this paper, the shape of the Somatosensory Evoked Potential (SEP) signals is analyzed to detect a focal lesion in the spinal cord of rats. SEPs can be used to reliably assess the spinal cord integrity. However traditional time-domain analysis often needs human intervention because of common difficulty in SEP peak detection due to low signal-to-noise ratio or flattening of SEPs during injury. We have proposed a novel shape analysis technique for SEPs, in which we obtain the slope information for the entire waveform in specific time bins. This technique does not involve any peak detection, and can be used without human intervention. We studied the effectiveness of this technique in a focal Experimental Autoimmune Encephalomyelitis model, in which a focal demyelinating lesion is induced in the spinal cord by injecting cytokine-ethidium bromide into dorsal white matter at T8 after immunization using Myelin Oligodendrocyte Glycoprotein combined with incomplete Freund's adjuvant. The technique was applied to SEPs recorded before and after lesion from 12 rats. The results showed the effect of injury to the SEPs from hindlimbs, hence demonstrating great potential of this technique to detect an injury to the spinal cord.

I. INTRODUCTION

Multiple sclerosis (MS) is an autoimmune disease involving destruction of oligodendrocytes responsible for creating and maintaining myelin. In focal experimental autoimmune encephalomyelitis (EAE), a rodent model of MS, a focal demyelinating lesion is induced by injecting inflammatory factors directly into the spinal cord of immunized rats [1].

Somatosensory evoked potential (SEP) is a reliable electrophysiological measure to assess the integrity of spinal cord [2]-[3]. It is the electrical response of nervous system to a sensory stimulus, recorded from the somatosensory cortex [4]. The traditional SEP analysis method [2]-[5] utilizes the amplitudes and latencies of few specific peaks. However these peaks are often obscured by noise and may even vanish in some injuries. This necessitates human intervention for the detection of peaks and hence, is often subject to errors.

This problem prompted us to develop a new technique [6] for quantifying the morphology of the SEPs as a whole, rather than a few salient peaks. This technique is largely borrowed from shape analysis tools in the image processing field [7].

II. METHODOLOGY

A. Materials & Methods 12 adult female Lewis rats with an average body weight of

200-220 gm were used. All experimental procedures were

approved by the Institutional Animal Care and Use Committee at Johns Hopkins University. All the surgical procedures were performed under anesthesia (mixed flow of 1.5% isofluorane, 80% oxygen and room air) delivered at 2 Llmin.

Four trans cranial screw electrodes (E363/20, Plastics One Inc.) were implanted on somatosensory cortex corresponding to both hind and forelimbs in each hemisphere. A fifth reference electrode was implanted on the frontal lobe. Subcutaneous needle electrodes (Safelead F-E3-48, Grass­Telefactor) were used to electrically stimulate the median and tibial nerves of both left and right limbs. An isolated constant current stimulator (DS3, Digitimer Ltd.) generated positive current pulses of 3.5 rnA magnitude and 200 J.lsec duration at a frequency of 1 Hz, which sequentially stimulated each of the four limbs at a frequency of 0.25 Hz using an advanced demultiplexer. SEPs from the trans cranial electrodes were recorded using an optically isolated biopotential amplifier (Opti-Amp 8002, Intelligent Hearing Systems) with a gain of 30,000 and a sampling frequency of 5 kHz.

Recombinant myelin oligodendrocyte glycoprotein (MOG) corresponding to the N-terminal sequence of rat MOG amino acids 1-125 was emulsified in incomplete Freund's adjuvant (Sigma-Aldrich) as 1:1 mixture. 50J.ll of this solution (Dr. Sha Mi, Biogen-Idec) was injected subcutaneously at near the base of the tail of each rat 2 contralateral sites for immunization. Prior to lesion, a 30 minute 'baseline' SEP was recorded. This was followed by a laminectomy at thoracic vertebra T9, where (2x2J.ll) cytokines (250ng ofTNF-a, 150U ofINF-y and 40ng ofIL-6) and ethidium bromide (lJ.lg) were injected into dorsal white matter using Hamilton needle (31G). A 30 minute 'post­injury' SEP was recorded after 7 days oflesion induction.

B. SEP Analysis All signal processing and statistical analysis was performed

using Matlab 7.0 (The Mathworks, Inc.). Contralateral SEP recordings were used for analysis. The signal-to-noise ratio was improved by ensemble averaging of 100 stimulus-locked sweeps, shifting the average window by 20 sweeps each time.

The slope information of signal x was expressed as the angular orientation () over a time segment f1t given by:

(J (X(t+f1t)-X(t»)

= arctan f1t

(1)

where f1t = 1/fs = 0.2 ms. We obtained this slope information for the entire signal, and clustered the angular orientations ()

This research was supported by the Maryland Stem Cell Research Fund (2007-MSCRFII-0159-00) and the Johns Hopkins Project RESTORE Fund.

into time bins of T = 2 ms and angle bins of e = 5°, thus constructing a 2-D slope histogram [6]. We also obtained mean slopes for each time bin:

1 N Omean = - I 0i (2)

N i=1

where N = T/!J.t = to. We obtained these mean slope values for all the rats from both the 'baseline' and 'post-injury' SEP recordings. We then tested for significant differences in absolute mean slopes between the two stages using a Student's paired sample two-tailed t-test.

III. RESULTS

Fig. 1 shows a sample SEP signal for both 'baseline ' and 'post-injury' recordings of a hindlimb of a single rat. The first 4 msec in an SEP waveform are generally corrupted by the stimulation signal, and do not give much relevant information. So this part of the waveform is excluded for our analysis.

The mean absolute slopes obtained from the averaged SEP (up to 28 msec) for all the rats, for both 'baseline' and 'post­injury' recordings, are shown in Fig. 2 in the form of grouped box plots, along with the significance codes. As it can be clearly seen, the forelimbs do not indicate apparent significant differences between the two stages. However in the case of hindlimbs, a potential significant SEP alteration between the two stages is evident. This indicates that hindlimbs are affected by the spinal cord injury.

Baseline Post-injury ~ 100 ,----------, ~ 100 I I I I

I I I I I ..... I I

:- I I I I I

I I I I

CI> -g 0 :!: Q.

.s ·100 10 20 30 40 50 '" 1 0 20 30 40 50

Time (msec) Time (msec)

Fig. 1. Sample ' baseline ' and ' post-injury ' SEP signals (50 sweeps) recorded from a rat's hindlimb

80

Ui" CI> ~ 60 Cl CI>

~ 40 CI>

Cl c: 20 <

0

80 Ui"

CI> CI> 60 a, CI>

~ 40 CI>

Cl c: <

4

Right Forelimb

1

l I 1 I I I I l I I , ,

6 8 10 12 14 16 18 20 22 24 26 28

Time (msec)

Right Hindlimb

6 8 10 12 14 16 18 20 22 24 26 28

Time (msec)

It is to be noted that the hindlimb SEP signals for the two stages are significantly different over a time period of 4-28 msec. This shows that the effect of spinal cord injury is not concentrated around the peaks alone, but rather over a sizeable portion of the waveform.

IV. CONCLUSION AND DISCUSSION

Peaks do not frame the entire signal length ofSEP. They do not reflect the full extent of the deformational change that occurs after injury. The use of slope changes gives us a more global representation of potential changes throughout the waveform. The proposed technique effectively differentiates between the two stages (before and after injury) based on mean slope changes over the entire SEP waveform, without any need of human intervention.

REFERENCES

[I) M. Kerschensteiner, et aI. , "Targeting Experimental Autoimmune Encephalomyelitis Lesions to a Predetermined Axonal Tract System Allows for Refined Behavioral Testing in an Animal Model of Multiple Sclerosis," Am. J. Pathol., vol. 164, pp. 1455-69, 2004.

[2) B.L. Grundy, "Intraoperative Monitoring of Sensory Evoked Potentials," in Evoked Potentials II, R.H. Nodar and C. Barber, Eds. Boston: Butterworth, 1984, pp. 624.

[3) M.R. Nuwer, "Spinal Cord Monitoring with Somatosensory Techniques," J. Clin. Neurophysiol. , vol. 15, pp. 183-93, 1998.

[4) G.D. Dawson, "Cerebral Response to Electrical Stimulation of Peripheral Nerve in Man," J. Neurol. Neurosurg. Psychiatry, vol. 10, pp. 137-40, 1947.

[5) E. EI-Negamy, and E.M. Sedgwick, "Properties of Spinal Somatosensory Evoked Potential Recorded in Man," J. Neurol. Neurosurg. Psychiatry, vol. 41 , pp. 762-68, 1978.

[6) G. Agrawal, D. Sherman, N. Thakor, and A. All, "A Novel Shape Analysis Technique for Somatosensory Evoked Potentials," Con! Proc. IEEE Eng. Med. BioI. Soc., pp. 4688-91 , 2008.

[7) F. Mahmoudi, 1. Shanbehzadeh, A.M.E. Moghadam, and H.S. Zadeh, "Image Retrieval Based on Shape Similarity by Edge Orientation Autocorrelogram," Pal/ern Recog. , vol. 36, pp. 1725-36, 2003.

80

Ui" CI>

60 ~ Cl CI>

~ 40 CI>

Cl c: 20 <

0

80 Ui"

CI> CI> 60 a, CI>

~ 40 CI>

Cl c: 20 <

4

Left Forelimb

,1

I: i I I I ' I I

]1 , j

II l i 6 8 10 12 14 16 18 20 22 24 26 28

Time (msec)

Left Hindlimb

6 8 10 12 14 16 18 20 22 24 26 28

Time (msec)

Fig. 2. Grouped box plots showing the mean absolute slopes for all the rats, for both ' baseline ' (cyan) and ' post-injury ' (magenta) stages. Asterisks indicate statistical significance of mean differences between the two stages using following codes for the p-values: *<0.05 **<0.01 ***<0.001.