Time-Frequency Analysis of Motor-Evoked Potential in Patients with Stroke vs Healthy Subjects: a Transcranial Magnetic Stimulation Study

  • Neha Singh
  • Megha Saini
  • Nand Kumar
  • K. K. Deepak
  • Sneh Anand
  • M. V. Padma Srivastava
  • Amit MehndirattaEmail author
Part of the following topical collections:
  1. Topical Collection on Medicine


Conventional analysis of motor-evoked potential (MEP) is performed in time domain using amplitude and latency, which encapsulates information relevant to the cortical excitability of the brain. The study investigated the importance of time-frequency analysis by comparing MEPs in time-frequency domains (TFD) of healthy versus stroke survivors. Six healthy subjects and ten patients with stroke were enrolled. Single-pulse transcranial magnetic stimulation (TMS) at resting motor threshold (RMT) was given at extensor digitorum communis muscle cortical representation to obtain MEP. MEPs were obtained at resting motor threshold (100% RMT subjects and patients), supra-threshold range (100–170% RMT), and different voluntary contractions (100% RMT) to subjects. Fast Fourier transform and continuous wavelet transform (CWT) were used for analysis. Frequency spectrum showed 98% and 66% of signal power in 0–100 Hz for subjects and patients, respectively. Top 10, top 25, and top 50 percentile power of CWT were calculated for each MEP. Frequency spectrum of top 10 and top 25 percentile power of subjects were different (p < 0.05) and dispersed to 0–500 Hz for patients; both groups having a 40-Hz peak. Total power of MEP was found to be low (p < 0.05) in patients as compared to subjects and top 10, top 25, and top 50 percentile power showed decrease. Clinical scores—MAS and FM—were observed to be correlated to frequency and time-frequency features (p < 0.05). Frequency spectrum belonging top 10 percentile power of different level voluntary contractions showed statistical significance (p < 0.05). However, no significant differences were observed for MEPs at different supra-threshold intensities. Results suggest time-frequency analysis might provide objective ways to quantify TMS measures for stroke patients.


Stroke Time domain features Time-frequency domain analysis Frequency spectrum Extensor digitorum communis muscle Modified Ashworth Scale 


Since the discovery of magnetic stimulation of the brain, transcranial magnetic stimulation (TMS) has been used in numerous clinical applications with its therapeutic potential being an active research area [1]. TMS has been shown to be a valuable tool in studying the regional localization [2], connectivity of brain [3], pathophysiology of neurological disorders [4], and diagnostic utility [5]. TMS technology has its advantage of being non-invasive, painless, good spatiotemporal resolution, and ability to control the amount of stimulation intensity to be delivered [6, 7].

TMS can be used either as single-pulse TMS (sTMS), dual pulse TMS, or repetitive TMS (rTMS) protocol. Motor-evoked potential (MEP) obtained using transcranial magnetic stimulation (TMS) can be recorded as electromyogram (EMG) activity in a target muscle. MEP encapsulates information relevant to the cortical excitability of the brain [1] providing insights into membrane excitability of neurons, conduction and functional integrity of cortico-spinal tract, and neuromuscular junctions and is of prognostic importance in disease monitoring [2].

Two important features of MEP in time domain are amplitude and latency. Variation in time domain features of MEP is currently used in clinics to evaluate disease status and measure treatment responsiveness [3, 4, 8]. Following single-pulse TMS, the pyramidal neurons are activated transsynaptically producing “I” indirect waves at the threshold and as the stimulation intensity increases, pyramidal axons are activated producing “D” direct waves [5]. The level of excitability of motor cortex decides the size of descending volleys and, hence, the amplitude of MEP [6, 7]. Conduction time taken by neural impulses to travel along the cortico-spinal projections to peripheral muscles is reflected in latency period of the MEP.

Quantitative measures of cortico-excitability in time domain amplitude and latency qualify as strong input candidates as a marker of excitability and inhibitory measures in different applications [1, 4]. In healthy subjects, MEP amplitude increases with an increase in stimulus intensity; amplitude tends to increase almost linearly with stimulus intensity reaching plateau phase at higher stimulus intensity. The curve is sigmoid-shaped in nature and referred as stimulus response curve (SRC) [7, 9, 10, 11].

Time domain (TD) features are sensitive to noise, are not always accurate and might often result in false-positive and false-negatives interpretations, and have irreproducibility [12, 13, 14, 15], limiting its wide acceptance in the clinical community. It usually evidences a steep learning curve requiring significant experience for which visual monitoring by trained technicians and clinicians are needed [12]. They cannot be used in the automated analysis [12]. Previous studies on various evoked potentials in animals [12, 16, 17, 18, 19, 20, 21] and humans [22] have demonstrated the importance of analysis in frequency domain (FD) and time-frequency domain (TFD) for different pathophysiology conditions and showed that TFD might have advantages over TD analysis. Applications of FD and TFD of MEP in stroke have not been reported in humans and animals.

In this study, we intended to perform TD, FD, and TFD analyses of MEP exploring the potential information which might provide improved quantitative neurophysiological parameters. In our study for MEP analysis, the commonly investigated forearm muscle extensor digitorum communis (EDC) in stroke has been chosen as also suggested in the literature [15, 23, 24]. The functional cortico-muscular connectivity and coherence, till date to the best of our knowledge, has only been studied using EEG, MEG, and EMG [25] but not using MEP which is a direct measurement of neuronal connections and relationship between cortical and muscular activity. We also explored time-frequency information of MEP, for healthy subjects and patients with stroke, driving muscle at different threshold intensities at resting states and different grades of isometric contraction.

Materials and Methods

Subjects and Electromyography Recording

Right-handed male healthy subjects (n = 6, age = 27.5 ± 7.2 years) with inclusion criteria age 18–70 years, no neurological-deficit/hypertension/diabetes, and compliant with TMS procedure were chosen for this study. Ten right-handed patients with stroke (n = 10, male/female = 8:2, age = 49.8 ± 14.85 years, Table 1) with the following inclusion criteria: within 24 months chronic, first, and unilateral ischemic/hemorrhagic stroke, age 18–70 years having no peripheral neuropathy, Modified Ashworth Scale (MAS) ≤ 3, compliant with TMS procedure were enrolled in this study. The clinical scales—MAS and upper limb Fugl-Meyer (FM) scale—were measured for all enrolled patients (Table 1). The MAS of wrist joint was assessed ranging from 1 to 3 on a scale of 0–4 and upper limb FM scale ranged from 20 to 52 on a scale of 66. The study was approved by the Institutional Review Board (IRB) at the All India Institute of Medical Science, New Delhi (IEC/NP-99/13.03.2015), for healthy subjects and patients with chronic stroke (more than 3 months to less than 24 months only). Written informed consent was obtained from all participants at the time of enrolment.
Table 1

Details of patients enrolled 


Age (years)/sex

Affected side


Chronicity (months)

MAS scales

Upper limb FM scale

Patient 1



R MCA, parietal lobe, basal ganglia ischemic




Patient 2



R MCA ischemic




Patient 3



R frontal ischemic




Patient 4



R pons + medulla ischemic




Patient 5



R thalamic + cerebellar hemorrhagic




Patient 6



R basal ganglia hemorrhagic




Patient 7



L parietal lobe ischemic




Patient 8



L gangliocapsular region ischemic




Patient 9



L temporoparietal and gangliocapsular ischemic




Patient 10



L basal ganglia hemorrhagic




R right, L left, MCA middle cerebral artery

The disposable gel-based wet Ag/AgCl surface electrodes were used in a bipolar configuration in which active electrodes were placed on the muscle belly with a center-to-center inter-electrode distance of 20 mm and ground electrode was placed on the lateral epicondyle. Muscle contraction causing extension of third digit of hand was observed for identification of muscle-belly and electrode placement. Electrodes were connected to the EMG amplifier connected with TMS (Magstim Rapid2, Magstim, UK). Sampling frequency of MEP was 6000 Hz, MEP was saved in .MRF (Meta Raster Format XML metadata) format for further processing.

Experimental Setup and Procedure

Subjects were positioned comfortably on TMS reclining chair in half supine posture, with left forearm pronated, elbow joint in 90–120° flexion, wrist joint in neutral position, and fingers at rest. MEP was acquired in a quiet place, the patients were instructed to close the eyes, keep the hand in a fully relaxed condition for 120 s before starting the experiment, and take deep breaths. The experiment was done at the same time of the day to ensure same experimental conditions for all the subjects. Specific hotspot for the EDC muscle was determined for each individual.

Single-pulse TMS were delivered to measure the MEP at cortical representation of the left EDC muscle (between Cz and C4 of contralateral primary motor cortex with reference to the EEG cap wore by them) for healthy subjects and left or right EDC muscle (between Cz and C4 or Cz and C3) for patients for the respective affected hand [26]. TMS stimuli were delivered by a flat 70 mm figure-of-eight coil (type D70 (AC), serial no. 0326, Magstim Rapid2, Magstim, UK), placed tangentially with handle pointing towards back, 90° to central sulcus and 45° to midsagittal line for trans-synaptical activation of the cortico-spinal tract [27]. TMS stimuli were delivered by moving the coil in millimeters in all directions until the hotspot was localized (“hotspot” is the area producing maximum MEP response for the respective muscle). Once the hotspot was localized, RMT was measured by progressively increasing the maximum stimulator output (MSO) starting from stimulus intensity of 35% in steps of 2 to 5% until a reliable MEP (>50 μV peak-to-peak) appears [7]. Then, MSO is lowered in steps of 1% until there are 5 consecutive responses out of 10 trials. Each pulse were given at an interval of > 5 s [28]. Once the RMT was determined for a reliable MEP at the hotspot, the stability of the hotspot of EDC muscle throughout the experiment was ensured by marking the area with a colored marker. We encountered approximation of hotspot in slightly lateral or posterior or lateroposterior as compared to unaffected side hotspot in the patient cohort in the study.

Healthy Subjects

MEPs were recorded by giving single-pulse TMS at the hotspot of EDC muscle of each healthy subject in three ways: (a) at 100% RMT, (b) at different intensity values (110%, 130%, 150%, and 170%) with increasing intensity expressed relative to the respective RMT in supra-threshold range, and (c) at isometric wrist extension at 50% MVC (maximum voluntary contraction) and 100% MVC at 100% RMT. Thus, in total, seven MEPs were recorded for each healthy subject. For the second way (b), the order of stimulus intensity was randomized for subjects avoiding anticipation bias in MEP recording. Also, the interval of 5 s was maintained in between each stimulus to avoid overlaps in refractory period of the last and action period of the next MEP. Once the hotspot was localized and RMT was determined, only the last MEP (5/10 trials) was saved (for each seven conditions for seven subjects) due to the hardware limitations of Rapid2 Magstim TMS as each MEP sample file needs to be saved manually and individually while maintaining the coil position at the hotspot.


RMT was determined for patients on their respective affected side by similar methodology as described above. MEP was recorded only at 100% RMT for patients, with single MEP for each patient.

Data Analysis

The sampling frequency of the MEP pod, built-in with TMS machine, is an effective 40-k samples per second. It is implemented using a 24-bit delta-sigma A-to-D converter with a 16-MHz, which gives an effective 40-k samples per second at 19 bits effective dynamic range. The second-order low pass anti-aliasing filter is fixed at 10 kHz. The first-order high pass filter to remove DC offset was set at 2 Hz. Electromagnetic interference (EMI) filters were also implemented. And then downsampling was done before storing the data at 6 kHz. Data analysis was performed using in-house build algorithms in MATLAB (R2013a, Mathworks, Inc., USA). Final data sample for analysis included the following: six MEP samples at 100% RMT for six healthy subjects (total six samples), four MEP samples for supra-threshold % RMT: 110%, 130%, 150%, and 170% each for six healthy subjects (total 24 samples), two MEP samples for MVC at 50% and 100% each for six healthy subjects (total 12 samples), and ten MEP samples at 100% RMT for ten patients with stroke (total ten samples). Hence, total 52 MEP samples were analyzed for each variable. The analysis window for all MEP samples was 49 ms that began after the TMS application to remove the artifacts. The analysis was performed on the whole 49 ms to remove the bias due to difference of latency and duration in healthy, stroke, and inter-variability within stroke population.

Frequency Domain Analysis

Frequency spectrum of MEP was analyzed using fast Fourier transform (FFT). The following features in FD were obtained: (a) power in the signal was calculated under different bands of frequency spectrum (0–100 Hz and 0–500 Hz) and (b) full width half maximum (FWHM) of the frequency spectrum of the signal.

Time-Frequency Domain Analysis

TFD analysis was performed using continuous wavelet transform (CWT). Grossmann and Morlet [12, 19] CWT was used as it is quite efficient in detecting lower amplitude signals and higher spectral components. Minimum time-bandwidth is provided by Morlet wavelets and signal is decomposed into a set of mother wavelets on the basis of functions like contractions, shifts, and dilations. CWT was calculated using FFT algorithm.
$$ \left[\mathrm{CWT}\left(t,a\right)=\sqrt{a}\int x{\left( a\tau \right)}^{\ast }{g}^{\cdot}\left(\frac{t-\tau }{a}\right){e}^{jt\omega}\right] $$
where a is scale factor for compression and expansion of the mother wavelet g(t), τ denotes time shift, and * is the complex conjugate. Function g(t) represents bandpass function centered around center frequency.

The following features in TFD were obtained: (a) magnitude of CWT coefficients; (b) total power of CWT coefficients in signal; (c) percentile power calculations—top 10 percentile, top 25 percentile, and top 50 percentile of CWT coefficients; and (d) range of specific frequency components in the signal at specific magnitudes. Higher magnitudes represented the higher contribution of the respective frequencies and vice versa.

Using MEP at 100% RMT, t test was used individually for each feature of TD, FD, and TFD comparing healthy subjects and patients for statistical significance (p < 0.05). t tests were also used for comparing the frequency spectrum of all supra-threshold intensities in healthy subjects evaluating differences. A one-way ANOVA test was used between groups 0%, 50%, and 100% MVC of the frequency content of top 10 percentile power determining the statistical difference (p < 0.05).

Pearson correlation coefficients of FWHM (in FD) and frequency content of top 10 and 25 CWT coefficients percentile power (in TFD) with two clinical scores (MAS and FM) were calculated individually to correlate these features with patients’ clinical conditions.


MEP was successfully recorded in six healthy subjects and 10 patients enrolled in the study. Clinical scores were successfully measured from all these patients (see Table 1 for details of patients). Tables 2 and 3 show various parameters obtained from time, frequency, and time-frequency domain of 6 healthy subjects and ten patients and their correlations with clinical scores.
Table 2

Mean (± SD) of time domain (TD), frequency domain (FD), and time-frequency domain (TFD) features



MEP amplitude (TD)

Latency (TD)

Power (0–100) Hz (FD)

Power (0–500) Hz (FD)

FWHM of signal (FD)

Total power (TFD)

Power (top 10 perc) (TFD)

Power (top 25 perc) (TFD)

Power (top 50 perc) (TFD)


55* (± 10)

186.4 (± 88)

16.5 (± 1.1)

82.89 (± 8.78)

96.5 (± 4.88)

52.78 (± 34.47)

1.31E+08* (± 9.8E+07)

57.76 (± 6.19)

91.51 (± 3.14)

99.62 (± 0.38)


73* (± 14)

124.5 (± 42.5)

22.5 (± 6.6)

66.5 (± 21.05)

92.4 (± 7.34)

52.26 (± 52.18)

4.15E+07* (± 3.8E+07)

55.62 (± 5.57)

88.05 (± 4.07)

98.93 (± 1.13)

*p < 0.05

Table 3

Pearson correlation coefficients (p value) of FD and TFD features with clinical scores and TD features of patients


Modified Ashworth Scale

Fugl-Meyer scale

Resting motor threshold

MEP amplitude

Full width half maximum (FD)


− 0.64


− 0.58

(p = 0.015)*

(p = 0.046)*

(p = 0.855)

(p = 0.072)

Maximum content of Frequency range of top 10 percentile power (TFD)


− 0.53


− 0.34

(p = 0.033)*

(p = 0.112)

(p = 0.903)

(p = 0.325)

Maximum content of Frequency range of top 25 percentile power (TFD)


− 0.36


− 0.35

(p = 0.025)*

(p = 0.29)

(p = 0.53)

(p = 0.31)

*p < 0.05

Time Domain Analysis

The mean MEP at 100% RMT for six healthy subjects was observed to be 186.44 μV peak to peak as compared to 124.51 μV for ten patients (Fig. 1a, b, Table 2, Appendix Table 4). SRC with relative supra-threshold stimulus (Fig. 1c) of MEP amplitude for healthy subjects showed the sigmoidal pattern of increasing MEP response with stimulus intensity. MEP increases sharply with stimulus intensity in 130–150% RMT, reaching a plateau with stimulus intensity > 150% RMT. Supra-threshold intensities were achieved by enhanced on option in TMS (present in Rapid2 Magstim). At 50% and 100% MVC, polyphasic response was observed with high amplitude (Fig. 1d).
Fig. 1

a Mean MEP response curve with 95% confidence interval (CI) for six healthy subjects at 100% RMT. b Mean MEP response curve with 95%CI for ten patients at 100% RMT. c Stimulus Response Curve (mean ± 95%CI) of six healthy subjects with the supra-threshold stimulus (at different % RMT) and MEP amplitude, showing the sigmoid shape curve. d MEP response curve at 50% MVC at 100% RMT

RMT for healthy subjects (55 ± 10) showed significant differences (p = 0.015) with lower values than patients (73 ± 14). MEP amplitude for healthy subjects (186.4 ± 88 μV) was considerably higher than patients (124.5 ± 42.5) (p = 0.078) and latency in healthy subjects (16.5 ± 1.1 ms) was considerably less as compared to patients (22.5 ± 6.6 ms) (p = 0.051).

Frequency Domain Analysis

In healthy subjects, frequency range of 0–100 Hz was observed with clear defined peaks of 40 and 60 Hz. In patients, the frequency spectrum was lower in magnitude with a spread in 0–250 Hz range, MEPs with multiple peaks of 40, 60, 80, 160, 200, 220, and 240 Hz, but highest peaks being at 40 and 60 Hz (Fig. 3a, b, Appendix Table 5). Table 2 compares the differences between TD, FD, and TFD features of MEP at 100% RMT for healthy subjects and patients. Power analysis of frequency spectrum at 100% RMT shows significant differences, out of full frequency range 0–3000 Hz, approximately 82% and 66% of power of signal lies in frequency range 0–100 Hz (p = 0.09). Also, though not significant, 96% and 92% of power of signal lies in frequency range 0–500 Hz (p = 0.24) in healthy subjects and patients, respectively (Table 2, Appendix Table 5). The FWHM of each MEP at 100% RMT of healthy subjects and patients were calculated (p = 0.971) (Table 2). Pearson correlation coefficients (CC) of FWHM was found to be positive with MAS (CC = 0.739 at p = 0.0145) and negative with FM (CC = − 0.64 at p = 0.046) (Table 3).

Figure 3a–d shows extending of oscillations towards higher frequency range with 0–100 Hz (peak frequency = 60 Hz) at 0% MVC to 0–200 Hz (multiple clearly defined peaks at 40 Hz, 80 Hz, and 140 Hz; peak frequency = 80 Hz) at 50% MVC and 0–500 Hz (multiple clearly defined peaks at 80 Hz, 140 Hz, 200 Hz, and 320 Hz; peak frequency = 140 Hz) with a spread at higher frequencies at 100% MVC (Appendix Table 5). Figure 3e shows the frequency spectrum of one representative healthy subject at supra-threshold stimulations. A defined pattern of frequency spectrum with peak frequency of 40 Hz was consistently observed at all stimulus intensities from 100% RMT to 170% RMT (Fig. 3e, Appendix Table 5), and increase in the magnitude of frequency spectrum was observed with increasing stimulus intensity similar to the SRC curve of Fig. 1c.

Time-Frequency Domain Analysis

Differences Between Healthy Subjects and Patients at 100% RMT

The presence of frequency range till 3000 Hz, shown in Fig. 4 in 0–500 Hz range, with varying magnitude has been evidenced during the full time interval of MEP (0–50 ms) for subjects and patients (Appendix Table 6). Though different peaks with fewer magnitudes were seen in the patients, the highest peaks were observed at 40 and 60 Hz in subjects and patients, similar to the frequency domain (Fig. 3a, b). Substantial differences in subjects and patients were observed in terms of contributing frequency range, 0–100 Hz in healthy subjects and 0–500 Hz in patients (Fig. 4a, b). Contributing frequency range for top 10 and top 25 percentile of CWT coefficients showed differences and dispersed to wider frequency range in patients as compared to subjects (p = 0.027, p = 0.022, respectively) but not many differences were observed in the frequency range of top 50 percentile showed (p = 0.059) (Fig. 5a, Appendix Table 6). Though the power in the contributing frequency ranges decreased in top 10, 25, and 50 percentiles (Table 2, Appendix Table 6), total power showed significant differences among subjects and patients (p = 0.023) (Table 2, Appendix Table 7).

The features—maximum content of frequency range of top 10 and top 25 percentile—were correlated with clinical scores; correlation coefficients with spasticity scale MAS was CC = 0.672, p = 0.033 and CC = 0.696, p = 0.025, respectively (Table 3). Though positive and negative correlations of these features were found with RMT and MEP, respectively, they did not show statistical significance.

On the evaluation of frequency variations during complete time-duration of MEP 0–50 ms (Appendix Table 8), peak frequency of 40 Hz was prominent during peak-to-peak MEP in subjects and patients and even more in supra-threshold intensities (more than 100% RMT) in subjects (high magnitude shown in red in Fig. 4a, b and Appendix Table 8).

Time-Frequency Analysis at Different MVC

The frequency spectrum was found to be much more dispersed for variation in 50% and 100% MVC as compared to 100% RMT at 0% MVC with frequency covering the range up to approximately 363 Hz (Fig. 4c, Appendix Table 9). A one-way ANOVA was applied and frequency range of top 10 percentile was found to be different (Fig. 5b). F value for the ANOVA test was 2784.16 with p = 1.1102E−16. All post hoc pairwise Tukey HSD comparisons suggest that the p value corresponding to the F statistic of one-way ANOVA is lower than 0.01 which strongly suggests that the frequencies are significantly different. Frequency spectrum were observed to be dispersed towards the right as % MVC is increased with clear defined multiple peaks shifting towards right too at 40 Hz, 80 Hz, and 140 Hz and peak frequency of 80 Hz and 140 Hz at 50% MVC and 100% MVC, respectively (Fig. 5b, Appendix Table 9).


The conventional interpretation of TD features is potentially distorted by noise and can result in false-positive and false-negative results given its limitation [12, 13, 14, 15]. The main target of this study was to determine if there are differences in spectra of stroke patients as compared to that of healthy subjects in TFD. Key results of this study are as follows: with 100% RMT and the healthy subjects present an energy distribution in MEP at a certain frequency range (0–100 Hz) in TFD. This energy in patients, with low magnitude, spreads over the range 0–500 Hz and the shift in location (latency) was also observed. Mean power in frequency range (0–100 Hz) and (0–500 Hz) was observed to be low in patients. The total power of CWT coefficients and frequency range contributing to top 10 and top 25 percentiles power of MEP signal showed statistically significant differences (p < 0.05) in patients. Mean power at different percentiles (10, 25, and 50) of CWT magnitudes were also observed to be low in patients (Table 2). The FD feature -FWHM and TFD feature- maximum frequency content in 10 and 25 percentile of CWT magnitudes were found to be statistically correlated with clinical scores of stroke patients (Table 3). The frequency range 20–70 Hz appeared in MEP might belong to piper rhythm in motor cortex.

Differences in Healthy Subjects and Patients in Time Domain

RMT was observed to be ~ 33% higher in patients (p = 0.015) as compared to healthy subjects. MEP amplitude was observed to be lower by ~ 50% and latency was higher by ~ 26% (Table 2) in patients, might be indicating loss of upper motor neurons (UMN) or affected cortico-spinal tract or peripheral nerves in stroke [2, 8, 29]. The trend of SRC (Fig. 1c) as has been observed for different muscles was also observed in frequency domain analysis (Fig. 3e); amplitude was found to increase with increase in stimulus intensity producing a higher MEP possibly because of accessing larger neuronal recruitment of cortico-spinal higher threshold motor neurons in form of D wave [10]. SRC reaches a plateau phase at ~ 150–170% RMT (Fig. 1c) as reported in the literature [7, 9, 10, 11], possibly being the maximum stimulus intensity for cortico-spinal neuronal firing and highest achievable neural recruitment possible [30]. None of the time domain parameters showed a correlation with clinical scores.
Fig. 2

Box and whisker plot of a resting motor threshold, b MEP amplitude, and c MEP latency for the healthy subjects and patients; RMT shows significant differences in healthy subjects and patients, but latency and MEP amplitude does not

Differences in Healthy Subjects and Patients in Frequency Domain

A considerable decrease in mean power in the range of 0–100 Hz was observed to be ~ 25% (82.89 to 66.5) and ~ 5% (96.5 to 92.4) in 0–500 Hz in patients as compared to the subjects. Results reflect that the power content present in 0–100 Hz in subjects is considerably dispersed over 0–500 Hz in patients (Table 2, Figs. 3a, b and 4a, b). The decrease in energy content and dispersion of power towards higher frequency range with pathology seen here is in line with TFD studies on evoked potential conducted on rat animal model [12, 19]. The FWHM was found to be positively correlated (with p = 0.01) with MAS (CC = 0.74) which might indicate towards the relationship between higher spasticity with large contraction of muscle fibers and more pathology with large dispersion (right sided) in frequency spectrum resulting in higher FWHM and vice versa, as confirmed in the animal model [16, 19]. The negative correlation (with p = 0.46) of upper limb FM scale with FWHM (CC = − 0.64) might reflect the decrease in the dispersion of frequency spectrum with the higher functionality of upper limb.
Fig. 3

Mean frequency spectrum (with 95%CI) of the MEP for six healthy subjects at a 0% MVC (resting state, 100% RMT). b Mean frequency spectrum (with 95%CI) of MEP curve for 10 patients (resting state, 100% RMT). c 50% MVC, d 100% MVC, and e supra-threshold intensities for one representative healthy subject

Fig. 4

Time-frequency domain analysis showing three-dimensional-view of magnitude of coefficients of CWT at100% RMT for representative a healthy subject (resting, 0% MVC), b patient (resting, 0% MVC), c % RMT for representative healthy subject at 100% MVC, and d healthy subject at 170% supra-threshold intensity

Differences in Healthy Subjects and Patients in the Time-Frequency Domain

TFD analysis shows distinct information about the spectrum of healthy subjects and patients. The features which showed statistically significant differences (p < 0.05) among healthy subjects and patients (Fig. 5a, Table 3) were (1) frequency spectrum of top 10 percentiles of CWT coefficients, (2) frequency spectrum of top 25 percentiles of CWT coefficients, (3) maximum frequency content of top 10 percentile of CWT coefficients, (4) maximum frequency content of top 25 percentiles of CWT coefficients, and (5) total power in MEP signal. These features in TFD qualify as the input candidates for differences between healthy subjects and patients. These findings showed that the changes in frequency range, in stroke patients as compared to healthy subjects, were statistically significant than the change in amplitude and latency in time domain showing it might be a more sensitive monitoring technique. The dispersion was shown in the frequency spectrum of top 50 percentile in full 0–3000 Hz range. A similar tendency of dispersion has been reported in rat models evaluating various evoked potential like MEP in spinal cord injury [16]. In patients, top 10, 25, and 50 percentiles power did show a considerable decrease (Table 2, Appendix Tables 6 and 7) and total power showed a statistically significant decrease in power by 216% (p = 0.023). The decrease might be attributed to the weak cortico-spinal tract or the reduced ability or loss of UMN.
Fig. 5

Box and whisker plot showing the comparison of frequency spectrum contribution at a top 10, 25, and 50 percentile (shown as prc) power of at 100% RMT frequency range between subjects and patients and b top 10 percentile at different grades of MVC at stimulus intensity of 100% RMT, found to be different (p < 0.05 represented by an asterisk)

The two features—maximum frequency content of a top 10 percentile of CWT coefficients and maximum frequency content of top 25 percentiles of CWT coefficients—of each patient were correlated with each patient’s spasticity scale and showed positive correlation with MAS. The tendency of frequency dispersion towards higher (right-side) frequency was in confirmation of the findings in rat animal model [16]. In humans, it could be explained by: the more is the pathology and the more is the spasticity, and the more is the contraction of muscle fibers, having a large working frequency, controlled by large motor neurons and, hence, the more is the dispersion of frequency [31, 32]. Thus, stroke leads to changes in time-frequency domain and can be a better indication of the pathophysiology of stroke than TD as monitoring in TFD offers the detection of time shifting, frequency shifting, frequency dispersion, power loss in contrast to amplitude, and latency in time domain.

The piper rhythm (20–70 Hz with 40 Hz peak), found in EMG and also in frequency coherence of EMG with MEG and EEG, in differing grades of voluntary contractions of muscles in healthy subjects has been reported in several studies [17, 25]. It has been shown to be driven by the contralateral motor cortex and plays a critical role in the pathophysiology of diseases [33]. The same frequency range with 40 Hz peak was also observed in our study in healthy subjects and patients with stroke (Fig. 3a, b, Appendix Table 5, 8). The range is more dispersed towards higher frequency with lower in magnitude in patients as compared to the healthy subjects. The presence of 40 Hz rhythm in upper limb forearm muscle might indicate the presence of piper rhythm in motor-evoked potential of stroke patients, the same range has been earlier reported in EMG signal from lower limb muscle in stroke patients and in other neurological disease like Parkinson disease [33, 34]. Fang et al. in 2010 showed low gamma band cortico-muscular coherence in poorly recovered stroke survivors with the help of EEG-EMG coherence [35]. This difference might be because of the methodological differences in the study (MEP and EEG-EMG coherence). As TMS stimulus is applied, D waves and I waves are recruited which sums up at anterior horn of the spinal cord and travels to the muscle for the motor fiber to fire [1, 6]. So, it is worth noting that MEP is more complex-signal that captures information on neuronal firing, cortico-spinal tract conduction, and muscle excitability that could also be reflected in the range and shift of frequency spectrum in our study. Also, these differences might be because of different chronicity, impairment, and lesion size and location.

As mentioned above, MEP is a combination of information initiating at the pyramidal axons (D waves), various synapses though neuronal circuits (I waves), conduction through the tract and muscle recruitment, limiting of higher frequency content (top 10 percentile of CWT coefficients) to ~ 100 Hz in healthy subjects and dispersed to ~ 500 Hz in patients highlights substantial fundamental alterations among healthy subjects and stroke patients. One example of difference might be the unorganized recruitment of higher motor fiber firing at higher frequencies even in the resting muscles. Even though most of the frequencies were present at the complete duration of MEP, the highest CWT magnitudes are contained by 40 Hz peak frequency lying in the period of peak-to-peak MEP (Appendix Table 8) representing maximum change during the peripheral evoke response. The frequency spectrum was observed to shift towards the right in time axis in patients (Fig. 4b) representing longer latency period and slower cortico-spinal tract conduction. The sample size is limited to interpret significant impact; a bigger study with similar analysis approach would be highly beneficial.

The frequency range captured by the electrodes in our study could contain the information of not just the muscle firing but the neuronal firing too, MEP is a complex signal, as the presence of frequency range up to 100 Hz in the brain has also been confirmed by electrocortiocography (ECoG) study, MEG, and other in vivo pyramidal cells studies [6, 17, 36]. This is the first study on humans to report 100 Hz frequency content in brain non-invasively using TMS.

Hence, stroke leads to changes in frequency and time-frequency domain features—FWHM, total power, maximum frequency content of top 10 and 25 percentile CWT magnitudes—and can be a better indication of pathophysiology of stroke than TD as monitoring in TFD offers the detection of time shifting, frequency shifting, frequency dispersion, power loss in contrast to amplitude, and latency in time domain. The application of time-frequency domain analysis might provide an alternative way to detect/diagnose and interpret results in automated MEP monitoring. This approach might be helpful for neurophysiologists by providing a practical improved method to quantify TMS measures by providing an advantage over traditional time domain analysis which could be worked out for real-time clinical applications in future research. This exploratory study opens a new pathway for automated monitoring of neurophysiological parameters of MEP in stroke.

Differences at Different Supra-threshold Intensities and Different MVC in Healthy Subjects

The increase in the magnitude of oscillations (as indicated in Fig. 1c by SRC curve, shown in Fig. 3e, Appendix Table 8) with increasing stimulus intensity (signal growing vertically) might indicate a relationship of particular frequency range with D waves being generated with TMS stimulus because D waves manifests the same trend of increasing magnitude with increasing stimulus intensity. The magnitude increases with frequency range with peak frequency being constant (Fig. 3e) indicating the higher cortical response due to increase in stimulus intensity but the same muscle recruitment (rate coding). This might represent the activation of higher threshold central and peripheral pyramidal neurons having faster propagations and producing larger action potentials which finally govern large motor units in the target muscles [1].

The oscillation in Fig. 3a–d extends, not shifts, towards higher frequency range with increasing peak frequency at 40 Hz, 80 Hz, and 120 Hz, following Henneman size principle [32] and indicating of firing of both slow and fast muscle fibers, with increasing MVC—0%, 50%, and 100%, respectively (Fig. 3a–d, Appendix Table 5, Figs. 4c and 5b, Appendix Table 9). The signal grows horizontally with increasing MVC and frequency range have not shifted but extended showing the involvement of slow muscle fibers too along with fast muscle fibers in line with size principle and rate coding. The differences in the frequency content of the top 10 percentile power of CWT at different MVC was found to be statistically significant at with p = 1.1102E−16, F value = 2784 (Fig. 5b). This increase in frequency range might be due to higher neuronal firing, as demand on the motor cortex is likely to be greater under these circumstances. Coherence analysis has revealed that some of the oscillations are transmitted, probably via the pyramidal tract, to the active muscles and may entrain them into the same rhythmicity. The spectrum was also observed to shift left in healthy subjects showing less latency (Fig. 4c) as MVC increases supporting the fact that motor neurons near threshold tend to discharge excitatory post-synaptic potential (EPSP) which summates giving short latency EMG [1]. For the different level of contractions, the cortico-muscular coherence is found to be in piper rhythm in literature.


One of the limitations of this study is the low sample size. It is worth noting that the acquisition of MEP is challenging in patients with stroke and this study is exploratory in nature evaluating the potential advantages of time-frequency analysis in MEP. Also, the age group of the healthy subject and patients with stroke was different which can be further explored in future studies as age affects the cortical circuits too. Patient population had a large amount of heterogeneity in terms of lesion location, chronicity, or even the effects of rehabilitation patients might be receiving since the episode of stroke. A larger study would get benefit with a time-frequency domain analysis methodology proposed in the study.


The study demonstrates the features of the time-frequency domain of MEP analysis for healthy subjects and patients that might be of clinical relevance in disease diagnosis or prognostic monitoring. It also indicates the presence of piper rhythm in healthy subjects and patients with stroke and changes in MEP frequency range during different MVC using non-invasive MEP with TMS stimulation.



The authors would like to express sincere gratitude to healthy subjects and patients who agreed to participate in the study. Also, they thank Mr. Vikas Kumar and Ms. Komal at TMS laboratory for their support during data acquisition and Mr. Dixit Sharma for the help in data analysis.

Authors’ Contribution

Conceptualization: NS, AM; data curation: NS,MS; formal analysis: NS, AM; funding acquisition: AM; methodology: NS, AM; resources: NK, SA, PS; supervision: AM; writing the original draft: NS, AM; writing the review and editing: NS, AM, KKD, NK, PS.


This work was supported by Science and Engineering Research Board (SERB), DST, Government of India (YSS/2015/000697). Neha Singh was supported with research fellowship funds from the Ministry of Human Resource and Development (MHRD), Government of India.

Compliance with Ethical Standards

Conflict of Interest

The authors declare that they have no competing interests.

Ethics Approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The study was approved by the Institutional Review Board (IRB) at the All India Institute of Medical Science, New Delhi (IEC/NP-99/13.03.2015).

Informed Consent

Informed consent was obtained from all individual participants included in the study.

Supplementary material

42399_2019_113_MOESM1_ESM.docx (70 kb)
ESM 1 (DOCX 69 kb)


  1. 1.
    Lefaucheur J-P, Andre-Obadia N, Antal A, Ayache SS, Baeken C, Benninger DH, et al. Evidence-based guidelines on the therapeutic use of repetitive transcranial magnetic stimulation (rTMS). Clin Neurophysiol. 2014;125:2150–206.CrossRefGoogle Scholar
  2. 2.
    Sliwinska MW, Vitello S, Devlin JT. Transcranial magnetic stimulation for investigating causal brain-behavioral relationships and their time course. JoVE (Journal of Visualized Experiments). 2014 Jul 18(89):e51735.Google Scholar
  3. 3.
  4. 4.
    Rossi S, Hallett M, Rossini PM, Pascual-Leone A, Avanzini G, Bestmann S, et al. Safety, ethical considerations, and application guidelines for the use of transcranial magnetic stimulation in clinical practice and research. Clin Neurophysiol. 2009;120:2008–39. Scholar
  5. 5.
    Chen R, Cros D, Curra A, Di Lazzaro V, Lefaucheur JP, Magistris MR, et al. The clinical diagnostic utility of transcranial magnetic stimulation: report of an IFCN committee. Clin Neurophysiol. 2008;119:504–32.CrossRefGoogle Scholar
  6. 6.
    Rossini PM, Barker AT, Berardelli A, Caramia MD, Caruso G, Cracco RQ, et al. Non-invasive electrical and magnetic stimulation of the brain, spinal cord and roots: basic principles and procedures for routine clinical application. Report of an IFCN committee. Electroencephalogr Clin Neurophysiol. 1994;91:79–92. Scholar
  7. 7.
    Rossini PM, Burke D, Chen R, Cohen LG, Daskalakis Z, Di Iorio R, et al. Non-invasive electrical and magnetic stimulation of the brain, spinal cord, roots and peripheral nerves: basic principles and procedures for routine clinical and research application: an updated report from an I.F.C.N. Committee. Clin Neurophysiol. 2015;126:1071–107. Scholar
  8. 8.
    Escudero JV, Sancho J, Bautista D, Escudero M, López-Trigo J. Prognostic value of motor evoked potential obtained by transcranial magnetic brain stimulation in motor function recovery in patients with acute ischemic stroke. Stroke. 1998;29:1854–9 Accessed 9 May 2018.CrossRefGoogle Scholar
  9. 9.
    Bestmann S, Krakauer JW. The uses and interpretations of the motor-evoked potential for understanding behaviour. Exp Brain Res. 2015;233:679–89. Scholar
  10. 10.
    Rothwell JC. Techniques and mechanisms of action of transcranial stimulation of the human motor cortex. J Neurosci Methods. 1997;74:113–22.CrossRefGoogle Scholar
  11. 11.
    Hess CW, Mills KR, Murray NM. Responses in small hand muscles from magnetic stimulation of the human brain. J Physiol. 1987;388:397–419 Accessed 2 Oct 2016.CrossRefGoogle Scholar
  12. 12.
    Zhang H, Oweis Y, Mozaffari-Naeini H, Venkatesha S, Thakor NV, Natarajan A. Continuous quantitative motor evoked potentials for spinal cord injury detection. 2nd Int IEEE EMBS Conf Neural Eng. 2005;2005:430–3.Google Scholar
  13. 13.
    Héroux ME, Taylor JL, Gandevia SC. Correction: the use and abuse of transcranial magnetic stimulation to modulate corticospinal excitability in humans. PLoS One. 2016;11:e0147890. Scholar
  14. 14.
    Grunhaus L, Polak D, Amiaz R, Dannon PN. Motor-evoked potential amplitudes elicited by transcranial magnetic stimulation do not differentiate between patients and normal controls. Int J Neuropsychopharmacol. 2003;6:371–8.CrossRefGoogle Scholar
  15. 15.
    Butler AJ, Kahn S, Wolf SL, Weiss P, Barker A, Beric A, et al. Finger extensor variability in TMS parameters among chronic stroke patients. J Neuroeng Rehabil. 2005;2:10. Scholar
  16. 16.
    Hu Y, Luk KD, Lu WW, Holmes A, Leong JC. Prevention of spinal cord injury with time-frequency analysis of evoked potentials: an experimental study. J Neurol Neurosurg Psychiatry. 2001;71:732–40 Accessed 16 Nov 2016.CrossRefGoogle Scholar
  17. 17.
    Braun JC, Hanley DF, Thakor NV. Detection of neurological injury using time-frequency analysis of the somatosensory evoked potential. Electroencephalogr Clin Neurophysiol Potentials Sect. 1996;100:310–8. Scholar
  18. 18.
    Wang Y, Zhang Z, Li X, Cui H, Xie X, Luk KD-K, et al. Usefulness of time–frequency patterns of somatosensory evoked potentials in identification of the location of spinal cord injury. J Clin Neurophysiol. 2015;32:341–5. Scholar
  19. 19.
    Zhang Z-G, Yang J-L, Chan S-C, Luk KD-K, Hu Y. Time-frequency component analysis of somatosensory evoked potentials in rats. Biomed Eng Online. 2009;8:4. Scholar
  20. 20.
    Wang Y, Li G, Luk KDK, Hu Y. Component analysis of somatosensory evoked potentials for identifying spinal cord injury location. Sci Rep. 2017;7:2351. Scholar
  21. 21.
    Wang Y, Cui H, Pu J, Luk KDK, Hu Y. Time-frequency patterns of somatosensory evoked potentials in predicting the location of spinal cord injury. Neurosci Lett. 2015;603:37–41. Scholar
  22. 22.
    Hu Y, Luk KDK, Lu WW, Leong JCY. Application of time-frequency analysis to somatosensory evoked potential for intraoperative spinal cord monitoring. J Neurol Neurosurg Psychiatry. 2003;74:82–7. Scholar
  23. 23.
    Wolf SL, Butler AJ, Campana GI, Parris TA, Struys DM, Weinstein SR, et al. Intra-subject reliability of parameters contributing to maps generated by transcranial magnetic stimulation in able-bodied adults. Clin Neurophysiol. 2004;115:1740–7. Scholar
  24. 24.
    Lazar RB. Principles of neurologic rehabilitation: McGraw-Hill, Health Professions Division; 1998. Accessed 10 Jul 2018
  25. 25.
    Grosse P, Cassidy MJ, Brown P. EEG-EMG, MEG-EMG and EMG-EMG frequency analysis: physiological principles and clinical applications. Clin Neurophysiol. 2002;113:1523–31.CrossRefGoogle Scholar
  26. 26.
    Homan RW, Herman J, Purdy P. Cerebral location of international 10–20 system electrode placement. Electroencephalogr Clin Neurophysiol. 1987;66:376–82.CrossRefGoogle Scholar
  27. 27.
    Werhahn KJ, Fong JKY, Meyer B-U, Priori A, Rothwell JC, Day BL, et al. The effect of magnetic coil orientation on the latency of surface EMG and single motor unit responses in the first dorsal interosseous muscle. Electroencephalogr Clin Neurophysiol Potentials Sect. 1994;93:138–46.CrossRefGoogle Scholar
  28. 28.
    Awiszus F. Chapter 2 TMS and threshold hunting. Suppl Clin Neurophysiol. 2003;56:13–23. Scholar
  29. 29.
    Kobayashi M, Pascual-Leone A. Transcranial magnetic stimulation in neurology. Lancet Neurol. 2003;2:145–56. Scholar
  30. 30.
    Day BL, Dressler D, Maertens de Noordhout A, Marsden CD, Nakashima K, Rothwell JC, et al. Electric and magnetic stimulation of human motor cortex: surface EMG and single motor unit responses. J Physiol. 1989;412:449–73. Scholar
  31. 31.
    Angelova S, Ribagin S, Raikova R, Veneva I. Power frequency spectrum analysis of surface EMG signals of upper limb muscles during elbow flexion—a comparison between healthy subjects and stroke survivors. J Electromyogr Kinesiol. 2018;38:7–16. Scholar
  32. 32.
    Henneman E. The size-principle: a deterministic output emerges from a set of probabilistic connections. Journal of experimental biology. 1985 Mar 1;115(1):105–12.Google Scholar
  33. 33.
    Brown P. Muscle sounds in Parkinson’s disease. Lancet. 1997;349:533–5. Scholar
  34. 34.
    Lodha N, Chen Y-T, McGuirk T, Fox EJ, Kautz SA, Christou EA, et al. EMG synchrony to assess impaired corticomotor control of locomotion after stroke. J Electromyogr Kinesiol. 2017;37:35–40. Scholar
  35. 35.
    Fang Y, Daly JJ, Sun J, Hvorat K, Fredrickson E, Pundik S, et al. Functional corticomuscular connection during reaching is weakened following stroke. Clin Neurophysiol. 2009;120:994–1002. Scholar
  36. 36.
    Rothwell J, Thompson P, Day B, Boyd S, Marsden C. Stimulation of the human motor cortex through the scalp. Exp Physiol. 1991;76:159–200.CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Neha Singh
    • 1
  • Megha Saini
    • 1
  • Nand Kumar
    • 2
  • K. K. Deepak
    • 3
  • Sneh Anand
    • 1
    • 4
  • M. V. Padma Srivastava
    • 5
  • Amit Mehndiratta
    • 1
    • 4
    Email author
  1. 1.Centre for Biomedical EngineeringIndian Institute of Technology Delhi (IITD)New DelhiIndia
  2. 2.Department of PsychiatryAll Indian Institute of Medical Sciences (AIIMS)New DelhiIndia
  3. 3.Department of PhysiologyAll Indian Institute of Medical Sciences (AIIMS)New DelhiIndia
  4. 4.Department of Biomedical EngineeringAll India Institute of Medical Sciences (AIIMS)New DelhiIndia
  5. 5.Department of NeurologyAll India Institute of Medical Sciences (AIIMS)New DelhiIndia

Personalised recommendations