Three Dimensional Innervation Zone Imaging in Spastic Muscles of Stroke Survivors

Abstract

Objective-

Outcome of botulinum toxin (BTX) therapy of post-stroke spasticity relies largely on accuracy of BTX injection to the proximity of innervation zones (IZs). Recently developed three-dimensional IZ imaging (3DIZI) is the only technique currently available to provide 3D distributions of IZs in vivo, yet its performance has not been validated under pathological conditions.

Approach-

The performance of 3DIZI was evaluated in the spastic biceps brachii muscles of four chronic stroke subjects. High-density surface electromyography (sEMG) and intramuscular electromyography (iEMG) were simultaneously recorded. The IZ location in the 3D space of the spastic biceps calculated using the 3DIZI technique from sEMG recordings were compared against the IZ location detected using intramuscular wires.

Main results-

3DIZI successfully reconstructed the IZs in the 3D space of the spastic biceps all four stroke subjects, with a localization error of 4.7±2.7 mm, and specifically a depth error of 1.8±0.4 mm.

Significance-

Results have demonstrated the robust performance of 3DIZI under pathological conditions, laying a solid foundation for clinical application of 3D source imaging in leading precise BTX injections for spasticity management.

1. Introduction

Spasticity is commonly recognized as a velocity-dependent hyper-excitability of the stretch reflex [1, 2]. It is a common complication of stroke that negatively impact the quality of life in patients, with a prevalence of approximately 20–25% in stroke survivors [3]. The pathophysiological mechanism underlying spasticity involves complex central, peripheral and muscular alterations [1, 4]. Botulinum toxin (BTX) is a chemodenervation agent that has been widely used as the first-line treatment for clinical management of focal post-stroke spasticity [5]. It blocks the release of acetylcholine at the neuromuscular junction pre-synaptically, and consequently paralyzing the spastic muscle [6]. Although BTX has been considered generally safe [7], questions has been raised regarding to the adverse effect such as muscle atrophy and loss of contractile tissue [8], atrophy of off-target muscles [8], varied treatment effectiveness [9] and high treatment cost [10].

Direct solution for these questions requires the execution of effective treatment with minimum injection dosage. Prior evidence has demonstrated that injection targeted at the motor endplate regions, indicated by innervation zone (IZ), can result in greater effects of muscle relaxation [6, 11]. It is therefore of great clinical significance to give reliable, objective injection guidance by in vivo characterization of the IZs of spastic muscles. Linear surface electromyography (sEMG) array has been mostly employed to non-invasively identify the IZ location by recognizing the reversal of signal polarity with a differential setup [12, 13]. However, sEMG array can only localize IZs along the muscle fiber direction. In cases of large muscles such as biceps brachii, it lacks proper estimation of IZ location in the mediolateral direction. More importantly, conventional techniques cannot localize IZs in a three-dimensional (3D) muscle space; specifically, the depth of IZs remains unrecognized.

Similar to employing EEG to reconstruct neural activities in the 3D brain space non-invasively in functional brain source imaging [14–20], high-density sEMG acquires the interference patterns of multiple motor unit action potentials (MUAPs) with a broad coverage, and therefore making possible the tracking of internal activities in deeper muscle regions; yet limited effort has been made to utilize high-density sEMG signals for functional muscle imaging [21–23]. It is only until recently that the technique has been employed to identify the 3D location of the IZs in vivo with decomposed MUAPs and electrically elicited M-wave recordings [24, 25].

However, as there is currently no in vivo technique available that can accurately image the IZ location in a 3D muscle space, it is difficult to validate the accuracy of sEMG source imaging approaches in human. Liu et al. demonstrated the performance of a decomposition-based 3D IZ imaging (3DIZI) approach using computational simulations, and further validated the accuracy of 3DIZI in the biceps brachii muscles of two healthy subjects with intramuscular recordings [24]. Despite the effort made, the applicability and performance of the 3DIZI technique under pathological conditions still remains unknown. Spastic muscles may suffer from anatomical, functional and neuromuscular alterations that could essentially affect the performance of IZ imaging. Assessing the performance of 3DIZI in patients holds great importance in broadening its applicability and laying out foundation for clinic translation. In this study, we examined the performance of 3DIZI technique in characterizing the IZ location of the spastic muscle of chronic stroke survivors.

2. Methods

2.1. Participants

In total four stroke subjects (one female and three males, mean age 52.8 ± 11.6 years) with diagnosed upper arm spasticity participated in this study. The average duration of the chronic stroke is 5.0 ± 4.2 years. Before the experiment, all subjects were well informed of the purpose and potential risks of the study, and gave written informed consent. The experimental protocol was approved by the University of Houston and University of Texas Health Science Center-Houston institutional review board.

2.2. Study protocol

The biceps brachii muscles of the paretic side were investigated. Subjects were seated in an experiment chair with their affected forearm secured in two adjustable metal plates (TRS-500, Transducer Techniques, Temecula, CA, USA) to perform isometric contractions with force feedback as in described our recent study [26]. The paretic forearm was secured to the neutral position and the contralateral arm rested symmetrically on a height-adjustable table. After skin preparation, two high-density (8 by 8 sensors) sEMG grids (TMSi, Enschede, The Netherlands) were placed adjacently over the skin of bicep muscle belly, giving eight axial columns of 16 sEMG sensors along the muscle fiber direction. Each sEMG sensor was 4.5 mm in diameter, and the inter-electrode distance (IED) was 8.5 mm. The location of two sEMG grids was adjusted to cover most of the biceps muscle area. A reference electrode was affixed on the lateral epicondyle of the humerus, and a ground electrode was attached to the wrist of the contralateral arm with a thoroughly soaked Velcro strap (TMSi, Enschede, The Netherlands). After the placement of surface EMG grid, a sterile bipolar fine wire electrode (VIASYS Healthcare, Madison WI) was inserted to the muscle belly for validation purpose. The wire electrode was inserted at an angle normal to the skin surface with the insertion site and depth noted. The hybrid EMG recording setup is exampled by figure 1.

Figure 1.

Placement of sEMG grids and intramuscular wire electrode

Patients were instructed to perform three trials of maximal voluntary elbow flexion with the attempt of largest peak torque defined as the maximal voluntary contraction (MVC) value. Then two trials of isometric elbow flexion at 10% MVC were performed for 10 s, with visual feedback from a monitor. sEMG signals were recorded via a Refa-136 amplifier (TMSi, Enschede, The Netherlands) at a sampling frequency of 2,048 Hz; while the intramuscular EMG (iEMG) were amplified and digitized at 2,000 Hz by a Bagnoli desktop EMG systems (Delsys, MA, United States). Signal quality was checked real-time through the recording computer. Trigger signals were sent for the synchronization of both recordings.

2.3. Signal processing

High-density monopolar sEMG signals were analyzed offline. Bad channels were manually identified and excluded from data analysis. Raw sEMG data was notch filtered at 60 Hz and bandpass filtered at 10–500 Hz with second order Butterworth filters. The contraction epochs were extracted for further signal decomposition. A k-mean clustering convolution kernel compensation (KmCKC) algorithm developed in our lab was employed in this study to decompose the high-density surface EMG signals into constituent MUAP trains [27]. Briefly, a k-means clustering techique was employed to properly select the firing instances of of individual motor units and estimate the initial pulse train. Then a modified multi-step iterative convolution kernel compensation (CKC) method was employed to update the estimated innervation pulse trains to improve the decomposition accuracy [27, 28].

Raw iEMG was notch filtered at 60 Hz and high-pass filtered at 3 Hz with second order Butterworth filters. The power spectrum density of a representative sEMG recording was estimated with FFT in signal epochs of 500 ms [29], shown in figure 2A. As the sampling rate is relatively low compared with standard iEMG decomposition requirement, no decomposition was performed [30]. The firing train of all motor units capatured by wire electrodes was identified by peak detection. The threhold was defined as either 5 times the mean absolute value or 20% of the maximal value of the signal [31], as exampled by figure 2B. The correlations between iEMG train and each decomposed MUAP train was assessed [32]. One matched intramsuclar firing instance (P_iEMG) was identified when one surface firing instance (P_sEMG) fell into its Δt range, where Δt was set to 2.5 ms. The correlation was defined by the ratio of matched PiEMG over the total P_iEMG.The decomposed MUAP train correlated best with the iEMG was selected, shown as in figure 2C. The corresponding 128-channel mapping of MUAP waveform was then reconstructed via a spike-triggered averaging technique and saved for further 3DIZI calculation [33, 34].

Figure 2.

(A) Normalized power spectrum density (PSD) of sEMG recording from a representative channel (B) Extraction of iEMG discharging instances. The green trace marks the threshold T defined for peak detection. (C) Comparison between the iEMG firing train and best correlated MUAP train decomposed from sEMG data

2.4. Modeling

A realistic computational model of the upper right arm was reconstructed from MRI axial images to implement 3DIZI calculation [25]. In brief, volume geometries were reconstructed from MRI and meshed into a finite element model, which consist of 234,103 tetrahedral elements and 40,865 nodes, using Abaqus 6.12 (SIMULIA, Providence, RI). The model comprises of the compact bone, cancellous bone, body fill, biceps brachii, triceps brachii, fat and skin, with conductivity values assigned respectively based on a previous report [24]. A source space of 34,619 evenly distributed current dipoles was used to solve the forward problem, with an inter-dipole distance of 2 mm. Axial muscle conductivity was assigned to be 5 times the conductivity of radial direction to achieve model anisotropy [35].

2.5. Inverse Calculation

The inverse calculation was implemented based on our previous study [24]. The linear relationship between the source vector J and the measurement vector Φ can be expressed as

$$\varphi =G\cdot J+n$$ (1)

where G is the lead field matrix that maps the sources space to the measurement space; n is the noise vector of the measurement space. An sEMG-informed weighted minimum norm estimates was employed to solve the inverse problem to the linear model above, expressed as:

$$\widehat{J}=R{\left(W^{T}W\right)}^{-1}{L}^T{\left(L\left(R{\left(W^{T}W\right)}^{-1}\right)L^T+\lambda I\right)}^{-1}\varphi$$ (2)

where $\widehat{J}$ represents the estimated strength of the current dipoles in the 3D muscle space; W is a weighted matrix which compensates for the undesired depth dependence; λ is the regularization parameter given by the L-curve method [24]. The source covariance matrix R contains of the prior knowledge of source distributions and was constructed by taking the advantage of the surface location of the IZs which were identified from bipolar MUAP mappings, exampled by figure 3A. A reference line from the identified IZ normal to the skin surface was defined and current dipoles within one IED, i.e. 8.5 mm, to the reference line was selected. The cylindrical-shape subspace formed (with a diameter of twice IED) was used to reconstruct spatial constraint on the source space in form of source covariance matrix R in equation (2). A higher weighting factor of 0.9 is assigned to the corresponding elements of R matrix for the dipoles inside the cylindrical subspace, and the active dipoles outside the subspace were penalized with a weighting factor of 0.1. The high density EMG profile when the IZ channel reached its peak value was extracted for further 3DIZI calculation.

Figure 3.

Examples of (A) IZ identification from bipolar MUAP mappings and (B) corresponding surface location of wire insertion and the identified IZ of the best correlated MUAP

2.6. Result validation

3D wire location was reconstructed in the upper arm model by the insertion site, angle and depth. The difference between the 3D wire location and the center of reconstructed active dipoles were calculated and defined as the localization error. The depth was defined as the distance towards the skin surface. Specifically, depth error was given by the difference between the depth of wire sensor and estimated source center.

3. Results

Two of the patients suffered from ischemic stroke (Subject#1, 4) and the rest two hemorrhagic stroke (Subject#2, 3). Two subjects had their left side affected (Subject#1, 3) and two had impaired motor function of the right arms (Subject#2, 4). Three of the four subjects were able to perform voluntary contraction of elbow flexor. One patient (Subject#3) with no elbow flexion but high resting muscle activity captured by sEMG grid was also included. Specifically for this subject, two 10-s EMG recordings were acquired yet no isometric contractions were performed.

An average of 11±5 MUs was decomposed. Figure 3B exhibits the overlapped location of identified IZ and wire insertion on an electrode grid mapping. Figure 4 shows two examples of reconstructed IZ clusters and wire location. The localization difference is summarized in table.1, with an overall average localization error of 4.7±2.7 mm, and specifically a depth error of 1.8±0.4 mm. There is no relation found between the type of stroke and reconstruction accuracy.

Figure 4.

Examples of 3DIZI results in two representative subjects (A) left arm and (B) right arm. Grey geometry shows the FEM model of the biceps brachii muscle. Black dots mark the location of sEMG sensors, red marks the location of wire sensor, and blue represents the final reconstructed sources correlating with the intramuscular wire electrode.

Table.1.

Comparison between 3DIZI imaging results and wire electrode location

Subject	Trial1		Trial2
	LE (mm)	DE (mm)	LE (mm)	DE (mm)
1	3.6	1.6	3.5	1.6
2	8.0	1.2	9.5	2.3
3	2.5	1.8	2.4	1.7
4	2.8	2.0	5.6	2.4
Overall Mean			4.7(2.7)	1.8(0.4)

LE: localization Error; DE: depth error;

4. Discussion

In this study, bipolar wire sensors were employed to validate the performance of 3DIZI. Intramuscular wire features highly localized uptake area [36], and the location of wire electrode therefore can be viewed as a direct indicator of IZ. To improve imaging accuracy, spatial constraint was used based on the prior knowledge of decomposed MUAP mapping, where the surface location of the IZ can be visually identified. In this study, an average localization error of 4.3 mm was observed and specifically a depth error of 1.8 mm was found. Results show marked imaging performance in the spastic muscle of four stroke subjects.

There are altered intrinsic properties of spastic-paretic muscles after stroke, including muscle atrophy [13], reduced muscle fascicle length [37], increased fat tissue [38], as well as complex neuromuscular alterations such as altered muscle innervation [39], disordered control of motor units and compromised motor unit activation [40]. Specifically, the anatomical alterations may affect the accuracy of 3DIZI where a standard model was applied. Second, the 3DIZI technique relies largely on the performance of sEMG decomposition, which is more challenging in the paretic muscles than the healthy ones [40]. At last, the ability of fine motor controls in patients to perform consistent forces was compromised. All these factors may negatively affect the performance of 3DIZI, necessitating the assessment of its performance in paretic muscles. Nonetheless, the resultant localization accuracy suggested the robustness of proposed 3DIZI approach in paretic conditions.

In supporting of clinical BTX injections, 3DIZI can be employed to image the IZ of each motor unit, forming clusters of active dipoles that deliver a representative sample of the global IZ location. Similarly, the surface IZ location can be identified from the MUAP mapping and used to inform 3DIZI calculation. The strong performance of 3DIZI in spastic muscles has also demonstrated its clinic prospect of leading precise BTX injections. It has been shown that injections 1 cm away from the IZ can result in up to 50% suppression of treatment effectiveness [6]; therefore it can be expected that providing the depth information of the IZ can further improve the spatial localization accuracy and, consequently the treatment effectiveness. The technique will possibly be beneficial for reducing the treatment cost, increasing treatment consistency and minimizing potential long term adverse effects.

sEMG decomposition technique was employed in this study to extract a high-density mapping of MUAPs. Raw sEMG is comprised of surface interference patterns, affected by signal superimposition and cancellation. Therefore source imaging results informed by EMG surface interference patterns are difficult to interpret in a physiological sense and may not be considered a good representation of IZ distributions. On the contrary, the MUAP mapping reconstructed from decomposed sEMG reflects the activity of single motor unit; as such, corresponding spatiotemporal imaging results can be used to track the spatial origin and propagation of MUAPs. Moreover, decomposed MUAPs can also facilitate estimation of surface IZ locations, by observing the signal symmetry at mediolateral direction and therefore avoid the overestimation of neighboring recordings [33]. 10% MVC was used in this study to secure a sparse MU firing captured by wire and surface sensors, as higher contraction forces may result in a denser firing instances and higher degree of signal superimposition which may introduce extra disruptions to signal decomposition and correlation. The performance of 3DIZI relies greatly on the decomposition performance and model accuracy. A model from subject-specific MR images is expected to markedly improve the 3DIZI accuracy, although being time consuming and expensive. In addition, other signal processing techniques such as spectrum interpolation and empirical mode decomposition can be employed for signal de-noising and conditioning [41, 42].

So far 3DIZI has only been tested in the biceps brachii, which features relatively large size and parallel fiber direction; yet it is expected that 3DIZI can be applied to estimate innervation in other muscles that are accessible from skin surface. As the fascicle arrangement varies in muscles, additional care should be taken to model the conductivity anisotropy. Despite some challenges in generalization, 3DIZI provides a potential 3D solution for reliable IZ reconstruction in vivo. It can also be employed to distinguish IZs for different myoelectric sources, where multiple muscles coexist.

The main limitation of the presented 3DIZI approach lies in the fact that elbow flexion is required; therefore it may not be applicable to the patient cohorts with absent or minimum elbow flexion. However, the imaging results from one patient with recognizable resting sEMG suggest a potential solution, by signal decomposition based on spontaneous MU activities of spastic muscles. To the best of our knowledge, this study presents the first effort to decompose sEMG signals and use it for 3D source imaging. However, only an average of 6 motor units (comparing to an overall average number of 11) were decomposed for this subject. Whether or not the limited number of decomposed motor units can be treated as a representative sample of the entire motor unit pool requires further exploration.

5. Conclusion

A validation study of 3DIZI technique was performed in the spastic muscle of four stroke survivors. The robustness of its imaging accuracy under pathological conditions has been demonstrated. The results suggest that 3DIZI can be a promising clinical guidance for precise injection of BTX.