Conduction Velocity Distribution Estimation using the Collision Technique Theory and Simulation Study

Swarna Sundar, Jose A. Gonzalez-Cueto*, Chad S. Gilbert

Department of Electrical & Computer Engineering Dalhousie University, Halifax, NS, Canada

Abstract: Current nerve conduction studies (NCS) are influenced by the activity of the largest active fibers, making it difficult to assess the state of smaller nerve fibers. This study is aimed at alternative diagnostic techniques for assessing carpal tunnel syndrome (CTS). A conduction velocity distribution (CVD) estimator based on the collision technique that incorporates volume conductor modeling is proposed and discussed in this paper. Simulations were run to evaluate the accuracy of the CVD estimator and compare its performance with previous CVD estimators based on the collision technique. Results show the improved accuracy of the proposed approach, which is able to provide estimates with a percent mean square error (PMSE) lower than 1.1% for all CTS cases studied and lower than 2% in the presence of additive white Gaussian noise. Simulations also showed that conduction slowing in the carpal tunnel (CT) segment is detected by the proposed technique and displayed as an increase in the number of low velocity fibers. Results suggest that both CVD estimator and amplitude parameter proposed can help detect the severity of CTS in a patient more accurately than current NCS.

Corresponding Author:

Keywords: Nonnegative least-squares estimation, volume conductor modeling, collision technique, nerve conduction studies, conduction velocity distribution.

1360 Barrington St, PO Box 1000, Stn Central Dptm Electrical & Computer Engineering, Halifax, NS, B3J 2X4, Canada E-mail: jose.gonzalez-cueto@dal.ca

1. Introduction Carpal Tunnel Syndrome (CTS) is a nerve conduction syndrome caused by localized compression of the median nerve at the wrist. CTS is generally detected using electrophysiological testing such as standard Nerve Conduction Studies (NCS). NCS are performed by placing stimulating electrodes at a distance from the recording electrodes and measuring the peak latency of the compound nerve action potential (CNAP) recorded. The sensory orthodromic response is generally of small amplitude and requires averaging [1]. Typical measurements made are amplitude, area under the waveform, latency and conduction velocity (CV). An average sensory CV of 45m/s or less indicates the presence of CTS [2][3]. NCS evaluate the function of large myelinated nerve fibers with the highest CV, i.e. Abeta fibers [2][4]. Selective evaluation of nerve fibers based on their diameter or CV is not feasible with these techniques. According to the literature [4], severity of CTS progresses from large nerve fibers to small nerve fibers. This is not a surprising finding given that NCS available are able to evaluate the largest fibers active at the time of recording. An early deficit in the activity of smaller nerve fibers will likely go unnoticed since the contribution of larger fibers to the CNAP recorded is significantly bigger than that coming from the smaller fibers [5]. Hence, a method to assess the diameter or CV of the active nerve fibers traversing the carpal tunnel will improve current CTS diagnostic techniques. Characterizing a nerve in terms of the probability density function (pdf) that describes the distribution of active fibers across a velocity interval can be done by estimating its CV distribution (CVD). If reliable CVD estimates for the median nerve fibers traversing the carpal tunnel are obtained this will be a useful parameter to describe the nerve fibers being affected in a CTS patient. A new method for estimating CVD is proposed in this paper. The estimation of the

electrical source needed for the CVD estimator makes use of a deconvolution approach applied to signals recorded using the collision technique as will be described in Section 3. The collision technique can be used to selectively activate nerve fibers of different diameters by varying the delay between two stimuli - a distal supramaximal stimulus and a delayed proximal stimulus [3]. There are two stimulation channels placed on the skin surface, one at the wrist (the distal) and another at the elbow (the proximal). The proximal and distal CNAPs are recorded using a bipolar channel consisting of two surface ring electrodes placed at the middle finger. The inter-stimulus interval (ISI) is the time interval between the delivery of the wrist stimulus pulse and the delivery of the elbow stimulus pulse. When the ISI is relatively large, the proximal CNAP does not collide with the distal CNAP, hence a response from all the nerve fibers activated by the elbow stimulus pulse is obtained. When the ISI is gradually decreased, the contribution from small nerve fibers reduces as the slow traveling action potentials generated at both stimulation sites start colliding and only the faster traveling action potentials get through to the recording electrodes placed on the finger (see Figure 1). <Figure 1> Of the action potentials getting through the one with the lowest CV is determined by the ISI value. This lowest velocity value is calculated as [3]:
CV = D + 3 .2 ISI 0.37 (Eq. 1)

where CV is the conduction velocity in m/s,

D is the distance in mm between the two stimulation sites, i.e. wrist & elbow, ISI is the time in ms between delivery of wrist and elbow stimulation pulses.

2. Signal Modeling; SSFAP and CNAP

The CNAP response is the raw data from which the CVD is estimated therefore the CNAP modeling is an essential part of the CVD estimation technique. The CNAP is made of the superposition of single fiber responses, or single fiber action potentials (SFAP). In the approach proposed here signals are recorded at the skin surface requiring a description of the surface SFAP (SSFAP). The SSFAP is given as the convolution of the electrical source with tissue filter impulse response functions [5][6]:

d SSFAP (t , v ) K1 s (t ) h t , v v where h(t , v) = t [(r / v )2 + t 2 ]3 / 2 is the tissue filter impulse response and s (t ) = Vm (t ) t is the electrical source, Vm (t ) is the fiber membrane potential as an AP is generated, and * is the convolution operator, K1 is a constant accounting for the electrical conductivities of the media, r is the distance from fiber axis to recording point, i.e. fiber depth, d is the distance traveled between stimulation and recording site, v is the CV of the action potential, and t is the time variable.

(Eq. 2)

CNAP signals picked up by electrodes were expressed as the superposition of the SSFAP contributions. The elicited activity has been modeled as a linear summation of the temporally dispersed SSFAPs. Further, the fibers can be grouped into delay bins and the amplitude contribution of each class is assumed to be proportional to number of fibers in that class. Hence, the CNAP can be written as [5]:

CNAP (t ) =

i =1 mi SSFAP(t, vi )
M

(Eq. 3)

where M is the total number of delay bins, mi is the number of active fibers in bin i, bin i, or class i, groups fibers within a small range of delays [ i / 2, i + / 2] , vi is the CV representative of fibers in bin i, i = d vi is the delay corresponding to a fiber with velocity vi , and

is the delay bin width.

3. CVD Estimator 3.1. Source Estimator
A description of the electrical source used by the CVD estimator is essential. This source differs from one individual to another. We estimate it by measuring the CNAP contribution of a particular velocity range of fibers from the subject under study. By subtracting consecutive CNAPs obtained as the ISI is progressively reduced the CNAP contribution from fibers belonging to a certain velocity interval [vlow , vhigh] can be isolated (see Figure 2). <Figure 2> Two elbow CNAPs containing the contribution of fibers with velocities greater than vlow = d ( i + / 2 ) , CNAP1, and greater than vhigh = d ( i / 2 ) , CNAP2, are obtained. This is achieved using the collision technique with ISI values corresponding to vlow and vhigh, as determined from Eq 1. The two CNAPs are subtracted to get the CNAP contribution of active fibers in the velocity range [vlow , vhigh ] . This CNAP difference can be written as:
v high

CNAP (t ) =

SSFAP(t, vi )
4

(Eq 4)

v i = v low

Substituting eq.2 into eq.4, the CNAP difference can be expressed as [7]: CNAP (t ) = K1 mi s (t ) hi (t ) where s(t) is the electrical source to be estimated mi is the number of active fibers in the velocity interval [vlow , vhigh ] , and hi (t ) represents the average tissue filter function over the velocity range [vlow , vhigh ] , 1 d given by hi (t ) = h ( t , vi ) . mi v = v vi i low
v high

(Eq 5)

(Eq. 6)

Since there is no a-priori knowledge of mi, one thousand tissue filter functions were used to calculate this averaged tissue filter function, each corresponding to a different velocity within the interval [vlow , vhigh ] as shown below:
hi (t ) = where 1 999 d h t , vk 1000 k = 0 v k

v k = k v + vlow and v = ( v high vlow ) 999 .

The spectrum of the CNAP difference is then found using the Fourier transform

CNAP ( j) = {CNAP (t )} ,

(Eq. 7)

where CNAP ( j) is the frequency domain representation of the CNAP difference, and

{} is the Fourier transform operator.

The averaged tissue filter function (Eq. 6) used to describe the CNAP difference in Eq.5 is also converted to its frequency domain representation Hi(j) using the Fourier transform. The reason for finding these two spectral representations is shown next. Applying the Fourier

transform operator to both sides of Eq.5 and using the convolution property of the Fourier transform we obtain: CNAP ( j) K1 mi [ S ( j) H i ( j)] (Eq. 8)

Since K1 & mi are constants introducing a scaling factor, the spectrum of the source is estimated dividing the spectrum of the CNAP difference by the spectrum of the corresponding tissue filter function: ( j) = CNAP ( j) S H i ( j)

(Eq. 9)

This estimate of the source spectrum is converted to the time domain by taking its inverse Fourier transform, thus the source is estimated as: ( j) (t ) = 1 S s

(Eq. 10)

3.2. Linear Least-Squares Estimator

(t ) , i.e. replacing s(t) by s (t ) in Eq.2, SSFAPs are computed Using this source estimate, s for all M delay bins uniformly dividing the delay interval corresponding to the 20-80m/s velocity interval. Assuming availability of a recorded CNAP, the relative number of active fibers per bin m (the CVD) can be estimated using least-squares (LS) estimation. The CNAP in eq.3 can be expressed in matrix notation as: cnap = SFAP m where cnap is a K 1 column vector composed of K time samples of CNAP(t), i.e. cnap = (Eq. 11)

[CNAP(t1 ), CNAP(t 2 ),..., CNAP(t k )]T ,

SFAP is a K M matrix whose i th column is the time-sampled signal SFAP (t i , vi ) ,

(t ) in Eq.2, and where the SFAP(t,vi) are found substituting the source estimate s

is an M 1 column vector containing the relative number of active fibers corresponding to each of the M bins, i.e. m = [m1 , m 2 ,..., m M ]T .

The objective of the CVD estimation is to find the relative number of active fibers m i in each bin i. Eq.11 can be viewed as a set of K equations and M unknowns (the vector m). The solution to the unconstrained LS minimization problem is given by: = ( SFAP T SFAP ) 1 SFAP T cnap m (Eq. 12)

Since number of fibers are positive quantities, a non-negative constrained LS

is an estimate of 0 is solved instead. It should be noted that m minimization problem with m estimated in this manner is in the delay the relative number of active fibers in a delay bin. The m domain. This delay distribution estimate is converted into the CVD estimate as described in [5]. The resulting CVD estimate is denoted by w.

4. Simulations
Several elbow-elicited SSFAPs were generated by using Eq. 2 and then superimposed to make the CNAPs. These included CNAPs resulting from collision when both elbow and wrist stimuli were set off using different ISI values, these are used for estimating the source as described in Section 3.1. There was also a full CNAP obtained when only the elbow site is stimulated, this is used for the CVD non-negative LS estimation as described in Section 3.2. The following parameter values were used. The distances d from the elbow stimulation site to anode and cathode of the recording site at the finger were 400mm and 420mm,

respectively. The distance D between the two stimulation sites at the elbow and the wrist was 180mm. The depth r of the median nerve below the recording electrodes was taken as 2mm. The source s(t) used to build the SFAPs follows the mathematical function described by Plonsey [8] and was time scaled as described in [5][6]. This is the waveform the source estimator

(t ) s (t ) when successfully estimated. This original source and a aims at reproducing, i.e. s
sample of its estimation are shown in Fig 4.

4.1. CNAP Difference

Two velocities, called vlow and vhigh were selected to be the lowest velocities not to be blocked in two separate CNAP recordings as explained in section 1. In order to obtain the CNAP contribution from nerve fibers in the [vlow, vhigh] velocity interval these two CNAPs were subtracted. It was found in simulations that a smaller interval will generally give a better source estimate. An interval with a width of 1 m/s will estimate the source within about 0.01-0.07 % mean square error, and an interval with width of 5 m/s will estimate the source within 1-2 % mean square error. An example of the CNAP contribution of nerve fibers within the velocity range of [48, 52] m/s is shown in Fig.3. <Figure 3>

4.2. Source Estimation

The spectrum of the CNAP difference, which represents the contribution of nerve fibers from vlow to vhigh was substituted into Eq.9 to obtain an estimate of the source spectrum. The (t ) was then found by taking the inverse Fourier transform of its spectrum as source estimate s (t ) is shown in Fig.4 where it can be compared to the indicated in Eq.10. A sample estimate s original source s(t) used to generate the CNAP data. The estimated source is subsequently used to find the CVD estimate by means of the nonnegative LS method mentioned in Section 3.2.

<Figure 4> For finding the Fourier transform and inverse Fourier transform, MATLAB functions fft and ifft were used. To avoid the effect of the division by zero or by small amplitude components in the averaged tissue filter frequency response, H i ( j ) , components having amplitudes smaller than a predefined threshold were made 1 before the division in Eq.9. After this division the

( j) were linearly interpolated. corresponding components of the source spectrum estimate S

4.3. Percent Mean Square Error

In order to measure the estimator performance, the estimation error was found using the percent mean square error (PMSE) given by mean ( w v _ pdf ) 2 max ( v _ pdf ) 2

PMSE =

} x 100%

(Eq. 13)

where w is the CVD estimate, and v_pdf is the actual velocity pdf used for generating the CNAPs in simulations.

4.4 Velocity Distribution Estimation

In order to assess the potential of the proposed technique this was compared with the traditional approach to CVD estimation using the collision technique [3][8]. The traditional approach uses the difference in the areas of CNAPs recorded for subsequent ISI values as an estimate of the number of active fibers belonging to the velocity interval determined by the two ISI values. A Gaussian distribution with mean of 48m/s and standard deviation of 5m/s was used to generate the CNAP data to test both approaches: the traditional and the one proposed here. <Figure 5> Figure 5 shows that the proposed method estimates the original Gaussian distribution more accurately than the traditional method, 0.44% vs. 5.40% error. The traditional approach
9

shows a heavier bias towards the high velocities suggesting the presence of more high velocity fibers than there are. This is because the CNAP areas contributed by high velocity fibers are larger than those contributed by low velocity fibers as described by volume conductor theory, see expression for h(t,v) in eq.2. Since this velocity dependence is only accounted for in the new method proposed, the traditional approach assigns more fibers to high velocity intervals for which larger CNAP areas are found. To further evaluate the performance of the CVD estimator, two more scenarios were simulated. The CVD estimator was tested to see if it performs differently for distributions with different means in one experiment, and for different standard deviations (SD) in another. To test the CVD estimator performance for different mean values, several simulations were run, holding the SD at a constant value, and changing the mean conduction velocity from 30 m/s, up to 60 m/s. In this experiment, the errors on the CVD estimates were slightly higher for low mean velocities as shown below in table 1. This is due to a significant number of fibers reaching the lower boundary of the estimation interval when the distribution mean is 30m/s. <Table 1> A similar simulation was run to test the accuracy of the CVD estimator for different SD. The SD of the distribution was increased by 3 from 3 up to 15 as shown in table 2. It was found that the error on the CVD estimate increased for the distributions with large standard deviation, this is again connected with the distribution challenging the boundaries of the estimation interval for standard deviations of 12 & 15m/s. <Table 2> The low dependence of the estimator on the mean of the CVD and the good performance of the estimator for any SD up to 10m/s both suggest that this method would be suitable for accurately estimating any abnormal CVDs including expected or unexpected CTS cases. 10

In order to assess the potential of the proposed technique for differentiating stages of CTS in a patient, CVDs used to generate the CNAPs were varied. According to the literature [7], the nerve fibers with higher CV tend to be most affected in early stages of CTS. Figure 6A shows CVDs used to represent different cases of CTS. <Figure 6> Case 1 represents the CVD used for a healthy subject. This was a normal distribution with no skewness. In subsequent curves the number of active fibers in the higher velocity range is reduced in order to indicate the presence of CTS, with Case 5 being the most severe case considered. This is represented by its skewness and lowest amplitude indicating that fewer fibers remain active at this point. Attempts of estimating each distribution were undertaken. Ten realizations of each case were run and the PMSE was recorded for each of them. Table 3 shows the average CVD estimator performance for each case. <Table 3> As can be observed, the performance index stays close or below 1% in all cases, offering a consistent performance and a significant improvement with respect to the 5.4% of the traditional approach.

4.5. Amplitude of CNAP Contribution

For the various cases of CVDs shown in Figure 6 CNAP contributions of different velocity intervals were obtained as described in Sections 3.1 & 4.1. The amplitude of these CNAP differences were plotted against the corresponding CV intervals as shown in Figure 7. Test CVD's in Fig. 6B are similar to those in Fig. 6A, except that they are skewed towards the high CVs, exhibiting a deficit of low velocity fibers instead of high velocity fibers.

11

Case 1 represents healthy subjects in Fig 7. The CNAP amplitude is highest for the 45m/s-50m/s range and decreases gradually outside this velocity range. Typically, when the severity of CTS increases large diameter nerve fibers are affected before small nerve fibers. It can be seen from Fig.7A that the CNAP amplitude curves become left-skewed for cases 2-5. This indicates that the CNAP amplitude for the higher velocity range of fibers starts decreasing sequentially. These results show that the amplitude parameter has the potential for distinguishing the different stages of CTS. Hence, the amplitude of the CNAP difference could be used as another parameter, along with the CVD estimate, to assess the severity of CTS in a patient. <Figure 7> Fig.7B represents the CNAP amplitude curves for the cases presented in Fig.6B, where small nerve fibers are affected before the large nerve fibers. It can be observed that the CNAP amplitude curves are right skewed from Case 6 to Case 8, which indicates that amplitudes corresponding to the nerve fibers with lower conduction velocity starts decreasing first followed by the higher conduction velocity nerve fibers. This shows that amplitude curve has the flexibility to distinguish the different cases of CTS, i.e. it has the potential to find whether the large nerve fibers are affected first or if the small ones are affected first. Compared to the amplitude of CNAP contributions, CVD estimates show more promising results with respect to the information provided regarding fibers having lower CV. From Fig.7A, it was difficult to observe a noticeable change in the amplitude of the fibers with lower CV as the action potentials measured from these ranges were weak and scattered. Also, it is difficult to measure experimentally the CNAP amplitude of nerve fibers with lower CV using non-invasive skin surface stimulation techniques. However, for the CVD estimator independent measurement of the CNAP contribution from the lower CV range is not directly required. Hence, the CVD

12

estimate not only offers an unbiased estimate of the relative number of active fibers but it also offers a better idea about the nature of the nerve fibers affected than the amplitude of CNAP difference, although the latter can be used as a secondary parameter to assess CTS.

4.6. Slowing Due to Demyelination of Nerve Fibers

In the presence of CTS propagation of APs along the nerve fibers traversing the tunnel is not only obstructed but it is partially slowed in many of the fibers. A model representing the slowing in the conduction velocity of nerve fibers was incorporated to evaluate the performance of the CVD estimator. Two parameters were incorporated, the length of the CT segment that the median nerve passes through and the number of fibers that will experience slowing as a result of demyelination. It was assumed that as the severity of CTS increases, a larger number of nerve fibers will be slowed due to demyelination. The simulation assumes that any fiber in the median nerve can conduct with two distinct velocities: the normal healthy velocity or the slowed velocity in the case of demyelination. To determine the slow velocity corresponding to any nerve fiber, the following relation was found using information from [9]:
1

V 2 V D = 2.3 M 6 .6 where VD is the CV of the demyelinated fiber, and and VM is the CV of the myelinated healthy fiber.

(Eq. 14)

Another assumption made in the slowing model is that the nerve fibers are only slowed for a portion of the CT, not the whole distance. It is unlikely that every single Schwann cell in the CT segment has failed. The length of demyelinated nerve fiber was assigned to each slowed nerve fiber as a random number, uniformly distributed between 0 mm and the length of the CT.

13

It was further assumed that the odds of a large nerve fiber experiencing slowing were larger than that of a smaller nerve fiber. To simulate this, a certain number of the smaller fibers were protected from the effects of slowing in the model. Results for different amounts of slowing can be seen in figure 8. Represented in each figure is a plot of the CVD in a healthy part of the median nerve (PDF), the CVD representing the slowed nerve fibers in the CT (Slow), and the results from the proposed estimator; the collision-based method (CVD Collision). Figure 8A shows the results for the simulation when it was run with 10% of fibers slowed, and Figure 8B shows the results for 30% slowing. <Figure 8> As expected the estimator exhibits a certain bias when estimating the CVD of a nerve in which there is conduction slowing. The reason for this is that the estimator assumes all nerve fibers are conducting APs at different but constant velocities uniformly throughout the entire nerve path. It can be observed how the CVD estimate no longer follows the CVD existing outside the CT segment but rather approaches the slowed pdf describing the CT segment. As the slowed percent increases to 30%, see figure 8B, the estimator shows an increase of low velocity fibers, a positive result since it reflects what is going on in the segment of interest: the tunnel.

4.7. SNR Evaluation

Simulations were run to evaluate the performance of the estimator when noise is added to the signals. The signal to noise ratio (SNR) was calculated by dividing the peak value of the unblocked elbow CNAP signal by the standard deviation of the noise. The performance of the CVD estimator was assessed for different noise powers using the PMSE index. Simulations were run 10 times for each SNR holding all variables constant for the different noise realizations. Table 4 shows the average performance obtained for each SNR.

14

<Table 4> The results in the table show that as more noise is added to the signals the performance of the estimator slightly deteriorates, although the PMSE values and results obtained are still satisfactory especially for SNR values found in experimental conditions. Experimentally the SNR associated with the collision technique is between 6 and 20. Considering this range the table shows that a PMSE of 0.2-2.2% would be expected in the estimation of the CVD using the proposed method.

5. Conclusions
In previous approaches to CVD estimation based on the collision technique [3] [8] the SSFAPs used to form the CNAP were considered independent of the conduction velocity. In other words, the SSFAP was considered to be the same for all velocity ranges. However, the response from a single fiber propagates some distance from the stimulation site to the recording site, thus it is dependent on fiber conduction velocity. Also, the SSFAP amplitude and waveshape is dependent on velocity as determined by studies using volume conductor modeling [5]. The CVD estimator proposed in this study uses a volume conductor model to represent the SSFAP such that this velocity dependence is accounted for. To the best of the authors knowledge this is the first attempt at incorporating a volume conductor model into a CVD estimator based on the collision technique. The CVD estimations obtained in simulations for all health cases studied were all made with a PMSE smaller than 1.1%. This shows that the CVD estimator can perform well regardless of the nature of the nerve fibers affected during different stages of CTS. Once noise is considered the CVD estimates are accurate within a 2.2% PMSE, which is still well below the 5.4% error displayed by traditional techniques without noise being added. The proposed technique can also

15

offer CVD estimates that indicates the presence of conduction slowing of nerve fibers in the carpal tunnel segment. As results showed, the higher the percent of fibers showing slowing, i.e. suffering from demyelination, the higher the number of fibers the estimator displays on the low velocity end. Even though experimental testing and validation of the proposed technique would be required, simulation results are encouraging. These suggest that with the use of the proposed CVD estimator and amplitude parameter the severity of CTS would be evaluated by observing which velocity range has a relatively reduced number of active fibers and amplitude respectively, or by matching the patients CVD estimate to CVD templates identifying the advancement of the disease. It has been reported that as CTS progresses in a patient different fiber sizes/velocities are affected, starting at the high velocity end and affecting fibers with lower velocities as it becomes more severe. This is an aspect that could be further tested with the availability of the technique proposed here and its use in CTS patients. For example, if only small diameter ranges (low CV range) have a relatively reduced number of active fibers, as compared to results from healthy subjects, it may indicate a case of CTS in a patient that would be overlooked by current NCS. Hence, this method could help better describe the type of nerve fibers affected in a patient suffering from CTS or other peripheral neuropathies than current NCS techniques.

16

References
[1] J. Daube, Nerve conduction studies, In MJ Aminoff (Ed.), Electrodiagnosis in Clinical Neurology, Churchill Livingston, New York, 265-306, 1986. [2] A. Nishimura, T. Ogura, H. Hase, A. Makinodam, T. Hojo, Y. Katsumi, K. Yagi, Y. Mikami, T. Kubo, Objective evaluation of sensory function in patients with carpal tunnel syndrome using current perception threshold, J of Orthopaedic Science, 8, 625-628, 2003. [3] N. Dalkilic, B. Yuruten, B. Ilhan, Somatosensory conduction velocity distribution of median nerve middle palmar digital component, Intern. J. Neuroscience, 114, 153-165, 2004. [4] A. Nishimura, T. Ogura, H. Hase, A. Makinodam, T. Hojo, Y. Katsumi, K. Yagi, Y. Mikami, T. Kubo, A correlative electrophysiologic study of nerve fiber involvement in carpal tunnel syndrome using current perception thresholds, Clinical Neurophysiology, 115, 1921-1924, 2004. [5] J.A. Gonzalez-Cueto, P.A. Parker, Deconvolution estimation of nerve conduction velocity distribution, IEEE Transactions on Biomedical Engineering, 49, No.2, 140-151, 2002. [6] J.A. Gonzalez-Cueto, An investigation of non-invasive techniques for the estimation of conduction velocity distributions in skeletal muscles and nerve bundles, Ph.D. dissertation, Dept. Elect. Comput. Eng., Univ. New Brunswick, Fredericton, Canada, 2001. [7] S. Sundar, Selective stimulation of nerve fibers for detecting the severity of Carpal Tunnel Syndrome, MASc Thesis, Dept. Elect. Comput. Eng., Dalhousie University, Canada, 2005. [8] M. R. Caccia, A. Salvaggio, E. Dezuanni, M. Osio, M. Bevilaqua, G. Norbiato, A. Mangoni, An electrophysiological method to assess the distribution of the sensory propagation velocity of the digital nerve in normal and diabetic subjects, Electroencephalography and Clinical Neurophysiology, 89, 88-94, 1993.

R1

[9] Ritchie, J. M. Physiology of axons. In: The Axon. Structure, Function and Pathophysiology, edited by S. G. Waxman, J. D. Kocsis, and P. K. Stys. New York: Oxford Univ. Press, 1995

Acknowledgements
This work has been made possible in part by NSERC grant No. 262282.

R2

Mean CV
30 m/s 36 m/s 42 m/s 48 m/s 54 m/s 60 m/s

PMSE(%)
2.47 0.55 0.69 0.63 0.18 0.35

Table 1. Percent mean square errors (PMSE) of the CVD estimates for the scenario with different CVD means

Standard Deviation
3 6 9 12 15

PMSE(%)
0.46 0.88 1.28 4.44 3.65

Table 2. Percent mean square errors (PMSE) resulting from the estimation of the CVD when the original CVD has different Standard Deviations

T1

Health Case
1 2 3 4 5

Average PMSE(%)
0.56 0.74 0.65 0.74 1.09

Table 3. Percent Mean Square errors (PMSE) resulting from the estimation of the CVD for each of the five CTS cases.

SNR

CVD Estimator Error (PMSE)

0.218 % 0.300 % 0.537 % 0.945 % 2.235 %

50 20 10 5

Table 4. Signal-to-Noise (SNR) performance of the CVD estimator.

T2

Stim2 Elbow Nerve Trunk

Collision

Stim1 Wrist v1 v2

Recording at Finger

v1 < v2
Figure 1. Stimulation-recording collision setup. Stimulation pulse 1 delivered at the wrist, stimulation pulse 2 delivered at the elbow site following a short delay (ISI). Collision of slow fibers, e.g. v1, happens between elbow and wrist. Action potentials from faster fibers, e.g. v2, travel successfully from elbow to finger recording site.

F1

Collision

CNAP

Present in both CNAP1 & CNAP2

VLOW

VHIGH

velocity

CNAP1 CNAP2

Figure 2. The CNAP difference (CNAP = CNAP1 CNAP2) contains the contribution from nerve fibers with velocities larger than VLOW and smaller than VHIGH.

F2

1.5

0.5 [Arbitrary Units]

-0.5

-1

-1.5 0

6 Time (ms)

10

12

Figure 3. CNAP contribution of fibers with conduction velocities in the range 48-52m/s

F3

Figure 4. Original source and estimated source from interval 48-52m/s

F4

Comparison of Traditional Collision vs Proposed Method

0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 -0.01 20 30 40 50 60 70 80 Traditional Collision Actual pdf Proposed Method

velocity [m/s]
Figure 5. CVD estimates obtained by the traditional collision technique (dot-dash) with a PMSE = 5.40% and obtained by the proposed method (solid line) with a PMSE = 0.44%

F5

Figure 6. A) CVDs representative of the healthy case and four possible CTS stages, B) Other CVD cases showing the opposite effect shown in A, this is a loss of low velocity fibers with respect to the normal healthy Case 1. From Case 6 to Case 8 the number of fibers is gradually reduced more significantly on the low velocity side.

F6

160

Case 1 (Normal Case)

CNAP amplitude [Arbitrary units] .

140 120 100 80 60 40 20 0 20-25

Case 6 Case 7 Case 8

25-30

30-35

35-40

40-45

45-50

50-55

55-60

60-65

65-70

Conduction velocity range (m/s)

Figure 7. A) CNAP amplitude of different CV ranges of nerve fibers obtained for the different cases of fiber distribution shown in Figure 6A, B) CNAP amplitude estimate for each of the fiber distribution cases shown in Figure 6B.

F7

Figure 8. CVD estimate in the case of A) 10% slowing, B) 30% slowing (solid line), compared with the carpal tunnel velocity pdf (dashed) and the velocity pdf outside carpal tunnel (dotted)

F8