Journal of Clinical Neurophysiology Publish Ahead of Print DOI: 10.1097/WNP.0000000000000167
Usefulness of time-frequency patterns of somatosensory evoked potentials in identification of the location of spinal cord injury Yazhou. Wang a#, Zhiguo. Zhang b#, Xiang Li a, Hongyan Cuic, Xiaobo Xiec,K. D. K. Luk a, Yong Hu a,c* a
Department of Orthopaedics and Traumatology, The University of Hong Kong, Pokfulam, Hong Kong, PR China b
c
Department of Electrical and Electronic Engineering, The University of Hong Kong
Institute of Biomedical Engineering, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, PR
# These two authors contribute equally to this paper
TE
D
China
C C
EP
*Corresponding Author: Dr Yong Hu Dept. of Orthopaedics and Traumatology, The University of Hong Kong Address: 12 Sandy Bay Road, Pokfulam, Hong Kong Email address: yhud@ hku.hk Tel: (852) 29740359; Fax: (852) 29740335
Conflict of interest statement: None declared
Acknowledgements
A
The study is supported by the General Research Fund of Research Grants Council of the Hong Kong
SAR, China (767511M) and National Natural Science Foundation of China (81301287).
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Abstract
2
Somatosensory evoked potentials (SEPs) have been widely used to monitor the neurological
3
integrity of the spinal cord during spinal surgery. However, the location of neurologic impairment
4
cannot be determined from SEPs. Previous studies imply that the time-frequency characteristics
5
of SEPs may reflect the location of the spinal cord injury (SCI). To validate the hypothesis that
6
time-frequency patterns of SEPs are associated with the location of neurologic deficits in the
7
spinal cord, we studied the time-frequency distributions (TFDs) of SEPs at different injury levels.
8
Twenty-four rats were equally divided into one normal group and three injury groups, in which
9
weight-drop contusions were delivered to the spinal cord of the rats at C4, C5, or C6 level
10
respectively. By comparing the time-frequency patterns of SEPs across groups, we found
11
significant differences in several time-frequency regions of interest (ROIs) in the TFDs of the
12
normal group and the injury groups. Importantly, the ROIs were different across injury groups,
13
suggesting that these ROIs could be specific to injury locations. The results imply that changes of
14
the time-frequency patterns of SEPs may be related to the location of the SCI.
TE
D
1
15
Keywords: Somatosensory evoked potentials, Time-frequency analysis, Spinal cord injury, in
17
Intraoperative monitoring
18 19
Background
EP
16
Somatosensory evoked potentials (SEPs) have been widely used in intraoperative monitoring
21
during spinal cord surgery [1, 2]. In current clinical practice, intraoperative SEPs can provide
22
warning signs of SCI without any information about the exact location of the SCI. If a definite SCI
23
location can be identified, the surgeon could quickly investigate the SCI to attempt immediate
24
emergency treatment.
C C
20
SEP waveform contains a series of temporal-domain components that convey useful
26
information about the physiological mechanisms, reflecting sequential activation of the neural
27
structures along the somatosensory pathway [3–6]. It has been reported that waveforms in the
28
SEP possess some type of information regarding the location of neurological lesions [4, 7]. In
29
recent years, joint time-frequency analysis of SEPs was suggested as an alternative but effective
30
indicator of the SCI, as it can provide earlier and more sensitive clues about neural injury than
31
time domain measurements [8–10]. A previous study [9] showed that there is a set of stable
32
time-frequency components (TFCs) in SEP signals. These SEP TFCs could be remarkably altered
33
when neurological deficits occur in the spinal cord, which highlights the usefulness of SEP TFCs in
34
intraoperative spinal cord monitoring [10, 11]. Researchers have also found that some detailed
35
TFCs in SEPs might have different origins in the spinal cord [9, 12]. Based on these findings, we
36
propose the hypothesis that the time-frequency patterns of SEP are associated with the location
A
25
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
37
of the SCI. If this hypothesis is true, different patterns of TFCs could be indicative of the SCI in
38
different locations, providing valuable information for SEP monitoring.
39
This study aims to verify our hypothesis that SCIs at different locations could result in
40
different time-frequency patterns in SEPs and demonstrate the feasibility of detecting the
41
location of SCI using the TFDs of SEPs.
42
Methods Materials
44
The animal experiments were performed on twenty-four adult Sprague-Dawley rats
45
weighting between 250 g and 380 g. These rats were divided into four groups, three experimental
46
groups and one control (normal) group, each containing six rats. Rats in the three experimental
47
groups were injured at the C4, C5, and C6 level of the spinal cord respectively.
D
43
Experimental Procedure
49
During the experiments, each rat was placed in a stereotaxic frame to keep its head stable
50
and anesthetized with isoflurane (2% for induction and 1.5% for maintenance). Two screw
51
electrodes were implanted on the rat’s primary somatosensory cortices of the fore limbs, and a
52
ground electrode on the parasagittal right frontal lobe.
TE
48
After the occipital and nuchal areas were shaved and prepared, a skin incision was made to
54
expose the laminae from C3 to C7 with removal of ligamentum flavum between them through a
55
posterior approach under microscopy. A laminectomy was performed at a certain level of the
56
cervical vertebra (C4, C5, and C6 for the three injury groups respectively) to expose the dorsal
57
surface of the spinal cord underneath, without opening the dura mater. After that, a contusion
58
was performed using the NYU-MASCIS impactor [13] by dropping a 10-g rod from a height of 25
59
mm onto the exposed dorsal surface at a certain spinal cord level. By the end of the
60
electrophysiological examination, animals were sacrifised by anesthesia and the site of
61
laminotomy and lesion in the spinal cord was rechecked through anotomy to ensure the SCI level
62
was correct.
C C SEP recording:
A
63
EP
53
64
SEP signals were collected from the rats in the normal state and 20 min after contusion, using
65
SEP signal recording equipment (YRKJ-A2004; Zhuhai yiruikeji Co., Ltd., Zhu Hai, China). A
66
constant stimulator was used to generate a 4.1-Hz square wave 0.1 ms in duration to stimulate
67
the median nerve along the forelimb of the rat. The stimulation intensity was selected to cause a
68
mild twitch of the forelimb.
69
Cortical SEPs were recorded at a sampling rate of 10 kHz, amplified 2000 times and bandpass
70
filtered from 20 to 2000 Hz. For each rat, a total of 200 SEP responses were averaged to achieve a
71
good signal-to-noise ratio.
72
Time-frequency Analysis
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Previous studies showed that short-time Fourier transform (STFT) is suitable for
74
time-frequency analysis of SEP signals [10, 11, 14]. In this study, STFT with a Hanning window was
75
applied to the averaged SEP signals to calculate their TFDs. Each TFD was calculated from 20 to
76
200 Hz in a step of 1 Hz [9]. The length of the Hanning window was 35 ms to provide a good
77
trade-off between time resolution and frequency resolution [11]. All the TFCs in the SEP TFDs
78
calculated by STFT were localized using the method in [11], which mainly presented main
79
components of SEP as showed in the second row of Figure 1. After removing the main peak
80
(named as ‘main TFC’) and change the scale of plot, some sub-components with lower peak
81
power can be seen in the third row of figure 1. The TFDs calculated by STFT were expressed as
82
z-scores to make the comparison of TFDs among different groups possible. The z-score of a vector
83
input x is
D
73
z = (x − µ ) / σ
85
where μ and σ are the mean and standard deviation of the population x. The TFDs were also
86
log-transformed in order to make the t-test applicable [24].
TE
84
Statistical Analysis
88
Further, a point-by-point comparison of TFDs was performed to investigate the difference in
89
TFDs between the normal group and the other groups. For each time-frequency point, a t-test
90
(two-tailed, assuming unequal variances) was used to determine whether the TFDs of each injury
91
group were significantly different from those of the normal group, resulting in three p-value maps
92
in the time-frequency domain. A set of regions of interest (ROIs) was defined based on
93
time-frequency regions showing significant differences (p < 0.05). Summary values of ROIs (mean
94
of all ROI pixels [15]) in each injury group were then compared with those in the normal group
95
using a t-test.
97 98
C C
96
EP
87
Results
Representative SEP waveforms from four groups are shown in Fig. 1 (first row). It can be
seen that SEP waveforms after SCI change substantially compared with the normal status. The
100
TFDs of these four waveforms calculated using STFT are also shown in Fig. 1 (second row). The
101
time-frequency localizations of all TFCs are marked with white crosses on the log-transformed
102
and normalized TFDs (Fig. 1, third row). By visual inspection, these TFDs are different from each
103
other.
A
99
104 105
Figure 1. Typical SEP waveforms, their STFT-based TFDs, log-transformed and normalized TFDs (white
106
crosses: the positions of TFCs). (a) Prior to a SCI; (b) After a SCI at C4 level; (c) After a SCI at C5 level; (d)
107
After a SCI at C6 level
108 109
The distribution patterns of TFCs were compared across the four groups. All the TFCs in SEPs
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
110
could be classified into two categories based on their power charateristics. The TFC with the
111
maximum power in each TFD of the SEP was named as ‘main TFC’. The power values of all main
112
TFCs were higher than 1.5 (z-score value). TFCs with power lower than 1.5 (absolute z-score value)
113
were defined as ‘sub-TFCs’. The distribution patterns of the main and sub-TFCs of SEPs were
114
compared across groups in the time-frequency domain. Before the SCI, the main TFCs of SEPs were mainly distributed in the time-frequency regions
116
of 10–20 ms and 35–50 Hz (Fig. 2 (a)), which is consistent with the characteristics of a typical
117
upper-limb SEP in adult rats [16]. After injury, the main TFCs of SEPs spreaded to other regions
118
and had wider distribution ranges in the time-frequency domain. As shown in Fig. 2 (a), the
119
distributions of the main TFCs of the three injury groups are mixed, which implies that the level of
120
the SCI may not have an obvious effect on the postion of the main TFC of SEP.
D
115
122 123
TE
121
Figure 2. Distributions of the main TFCs and sub-TFCs of each group in the time-frequency domain. (a) Main TFCs; (b) All the sub-TFCs from 20-200Hz; (c) Sub-TFCs lower than 60Hz
124
Distributions of the sub-TFCs of the normal, C4, C5, and C6 groups are shown in Fig. 2 (b). In
126
the present study, the sub-TFCs between 60-200 Hz did not show a visibly discriminable pattern
127
among different groups. We could extract discriminable regions among the three injury groups in
128
the distribution area of the sub-TFCs lower than 60Hz. All sub-TFCs from 20Hz to 200Hz was
129
shown in Fig. 2 (b), while Fig. 2 (c) showed the detailed distribution of sub-TFCs lower than 60Hz.
130
It could be found that sub-TFCs in normal SEPs were mainly distributed in the temporal range,
131
from 30 to 50 ms. But sub-TFCs in the post-injury groups tended to have lower frequencies
132
(20–50 Hz) compared with the normal group (35–60 Hz). Half (8/16) of the post-injury sub-TFCs
133
were distributed in the area lower than 35 Hz, where there were no sub-TFCs in the normal group.
134
In addition, sub-TFCs in different post-SCI groups were located in distinct regions. Sub-TFCs in the
135
C4 group laid in the areas from 45 to 60 Hz and from 20 to 30 Hz. Sub-TFCs in the C6 group
136
concentrated in the region from 25 to 35 ms in the time domain and from 25 to 45 Hz in the
A
C C
EP
125
137
frequency domain. The sub-TFCs in the C5 group had a more scattered distribution compared
138
with the other groups: they could be found from 15 to 50 ms in the time domain and from 25 to
139
50 Hz in the frequency domain.
140
Further, the time-frequency maps of P-values calculated using t-tests showed that there
141
were a number of significantly different time-frequency pixels (p