Subjects were recruited from primary rehabilitation centers in Hong Kong. After recruitment, they were invited to the lab on the campus for the training. The inclusion criteria were as follows: (1) diagnosed as stroke with hemiplegia; (2) more than 6 months duration after the onset; (3) significant gait deficit (Functional Ambulatory Category, FAC < 4). The exclusion criteria were as follows: (1) any additional medical or psychological condition that would affect their ability to comply with the study protocol, e.g., a significant orthopedic or chronic pain condition, major poststroke depression, epilepsy, and artificial cardiac pacemaker/joint; (2) severe hip, knee, or ankle contracture that would preclude passive range of motion of the leg. All of the subjects gave written consent before the experimental procedures. This study was approved by the Joint Chinese University of Hong Kong-New Territories East Cluster (CUHK-NTEC) Clinical Research Ethics Committee (Ref. no: 2016.093), and all the procedures of this study have been operated in accordance with the Declaration of Helsinki. This study was registered on ClinicalTrials.gov (NCT 03208439). This study utilized a self-controlled design with the combination of clinical scales, EIM, and sEMG assessments on lower extremities after FES-assisted cycling training.

All the variables of EIM and sEMG were exported for offline analysis. Raw impedance variables were R, X, and θ values across multiple frequencies of alternating current from 5 to 1,000 kHz. Raw EIM data collected at the frequency of 50 kHz was analyzed and compared (28) because the frequency of 50 kHz has been widely used as a standard for single-frequency analysis in EIM (35). The AR of R, X, and θ (AR of R, X, and θ) were computed based on the raw variables.

The AR of X, R, and θ was computed by the following Equation (1):

where V represents the X, R, and θ values, respectively; V_Perp and V_Para represent these variables collected from perpendicularly and parallelly arranged electrodes, respectively.

The RMS and CI were used to characterize muscle activation and coordination.

We calculated the normalized RMS value by Equation (2):

Where i is the assessed muscle, including QC, HS, TA, and MG. RMS_i is the normalized RMS value of muscle i in one revolution. T is the length of the signal. A decrease in the RMS value represented the improvement of muscle activation volume, which means the subjects can perform the same task with lower efforts.

The CI value was calculated by the following Equation (3):

Where A_ij is the overlapping area of sEMG linear envelops for the muscles i and j in the warm-up cycling phase, and T is the length of the signal. The CI value varied from 0 (non-overlapping of the muscle pair activation in a specific movement) to 1 (the two muscles' activation totally overlapped with both normalized sEMG activation levels kept at 1 during the movement) (36). The reduction of CI after intervention revealed a more coordinated muscle contraction pair during cycling. Only CI values of agonist and antagonist muscle pairs (TA and MG, QC and HS) were calculated in the current study, to precisely analyze muscle coordination within one joint movement.

The clinical scores (FMALE, FMAac, FMAkc, 6MWT, and BBS), EIM variables (AR, R, X, and θ), and sEMG parameters (CI and RMS) were reported using mean and standard error. The “1+” score in MAS was coded as “1.5” for statistics (37, 38). Comparisons of EIM parameters between paretic and nonparetic sides as well as before and after all the training were conducted by the paired t-test if the data set was normally distributed. Comparison of clinical scores and sEMG variables between pretraining and posttraining was conducted by the paired t-test for the normally distributed data set. Data normality was verified by the Shapiro–Wilk test. Wilcoxon signed ranks test was performed for non-normally distributed data set (28). Pearson correlation was conducted on EIM, sEMG, and clinical scores if the data sets were normally distributed; otherwise, the Spearman correlation was applied. Statistical significance was defined as P <0.05, two-tailed for all calculations. Statistical analysis was conducted by SPSS23 (IBM Inc., WA, USA).

The EIM variables of R, X, θ, and AR of four measured muscle groups were compared between the paretic and non-paretic sides. The θ value of paretic muscles were significantly lower than that of the non-paretic side (QC: P = 0.004, TA: P = 0.009, HS: P = 0.002, MG: P = 0.004). The X values were significantly lower in the paretic muscles of QC (P < 0.001), TA (P = 0.007), and MG (P = 0.002). The R values were significantly lower in the paretic muscles including TA (P = 0.014) and MG (P = 0.027) (Table 2). After 20 training sessions, statistical results revealed that the θ value significantly increased in paretic TA (P = 0.014, pretraining: 12.6 ± 1.22, post-training: 13.64 ± 1.08), and MG (P = 0.017, pretraining: 11.04 ± 1.56, posttraining: 12.14 ± 1.68) (Figure 3). AR value comparison showed that only paretic TA presented a significant increase in the AR of θ value (P = 0.041, pretraining: 1.283 ± 0.062, posttraining: 1.339 ± 0.069) and AR of X value (P = 0.004, pretraining: 1.275 ± 0.054, posttraining: 1.374 ± 0.046) after training (Figure 4). QC and HS showed no significant alterations of EIM parameters after training (P > 0.05).

For sEMG parameters, RMS value of paretic TA significantly decreased after training (P = 0.022, pretraining: 0.24 ± 0.01, posttraining: 0.2 ± 0.01). The CI between paretic TA and MG significantly reduced after the training (P < 0.001, pr-training: 0.21 ± 0.01, posttraining: 0.16 ± 0.01) (Figure 5). Figure 6 demonstrates muscle contraction pattern alteration during three revolutions of cycling of one recruited subject, in which the contraction of TA and MG was more coordinated after training with less overlapping area under the curves (CI value of TA and MR muscle pair: pretraining: 0.168, posttraining: 0.139) (Figure 6). QC and HS showed no significant alteration of RMS and CI value after training (P > 0.05).

Significant increases in clinical scores (FMA-LE: P = 0.013, FMAac: P = 0.02, 6MWT: P = 0.009, BBS: P =0.038) (Table 1) were observed after the training. Mean MAS scores decreased after training, while the differences were not significant (Ps > 0.05). There was no significant alteration of FMAkc score after training (P = 0.058) (Figure 7). Significant correlation was demonstrated between EIM parameters and clinical scores, including AR of X value for TA and clinical score (AR-X-TA and FMA-LE score: r = 0.54, P = 0.046, AR-X-TA and FMAac score: r = 0.594, P = 0.025), as well as the X value of TA and clinical scores (TA-X and FMAac: r = 0.692, P = 0.006; TA-X and BBS: r = 0.628, P = 0.016). Significant correlation was indicated between sEMG and clinical scores, including RMS of TA and BBS score (r = −0.582, P = 0.029), RMS of MG and FMA-LE (r = −0.618, P = 0.019), and RMS of MG and FMAac (r = −0.653, P = 0.011) (Figure 7).

All paretic muscles presented lower θ values compared to the non-paretic muscles. The X value was reduced only in paretic QC, HS, and TA compared to the non-paretic side. The R value was decreased only in TA and MG. These outcomes were consistent with the previous study by Li et al. which reported that θ can be a more sensitive biomarker than X and R values in assessing the inherent muscle property of Biceps Brachii after stroke (28). We also observed significant lower AR values in paretic TA and MG. Reduced AR of X value and θ value after stroke was related to the loss of muscle fibers, increased fat, and connective tissue infiltration (24, 25). Previous histological studies also revealed muscle atrophy and fat tissue accumulation after stroke (12). This cross-sectional result supported that the EIM assessment could also be applied to muscle composition and structure evaluation of lower extremity muscles in chronic stroke survivors.

Previous studies reported that the increase in the CSA of muscle, muscle fiber hypertrophy, and intramuscular fat consumption were considered to be a longstanding effect after stroke rehabilitation intervention. Increased muscle mass and decreased intramuscular fat were related to the improvement of muscle strength (16, 39). Ryan et al. and Chae et al. found that muscle CSA and muscle fiber size increased and the intramuscular fat reduced after FES and cycling training (5, 40). Walls et al. and Herrero et al. found that the muscle fiber size and muscle CSA increased after FES intervention (41, 42). Natsume et al. (43) reported that FES could result in muscle hypertrophy because of recruiting both slow- and fast-twitch muscle fibers (43). Therefore, due to these factors, muscle fiber size increased and myofiber cytomembrane area enlarged, which increased the charge storage capability of the cytomembrane (22). This led to a higher time shift in this study when alternating current passed through the muscle in EIM evaluation, and resulted in the elevation of θ value after training (Figure 3). On the other hand, the AR value of TA increased after training in the current study, which further suggested a more regular myofiber arrangement after training compared to the one before training. In a fixed muscle volume, the increase of muscle fiber size caused the myofibers to arrange more tightly, which improved the characteristic of muscle anisotropy and made the electrical current flow tougher in the direction perpendicular to the fiber than in the longitudinal direction (29). In addition, the isotropic intramuscular fat reduction would increase muscle anisotropy property in return. These two points might be the reason for the increase in AR value after the training (Figure 4).

From a perspective of myofiber structure, ultrasonography was widely applied to evaluate muscle structure alteration after training. Liu et al. used B-mode ultrasound to assess TA and MG architecture alteration after 3 weeks of body weight-supported treadmill training in stroke survivors (44). Significant increases in pennation angle and muscle thickness after the training were reported in FES- assisted lower extremity training (44–47). In our previous cross-sectional study, EIM and ultrasound were jointly applied to explain muscle composition and structure alteration during passive muscle stretch for subacute stroke survivors. In paretic TA, the lower pennation angle and muscle thickness value might contribute to the reduction of the X and θ values compared to the non-paretic TA muscle (29). Therefore, in the current study, the increase of X and θ values on TA could be related to the higher pennation angle and muscle thickness value after FES-assisted cycling training. The elevation of the pennation angle suggested that more muscle fibers and contractile materials are attached to the tendon and contributed to more muscle fibers packed in the same cross-sectional area, which promoted force transfer to the tendon (48). Greater pennation angle and higher muscle thickness contributed to higher AR value because under these conditions, muscle fibers are arranged more perpendicularly, and the current would transversely flow through more muscle fibers in a fixed distance of two voltage electrodes during EIM assessment; ultimately, the AR value increased.

Functional electrical stimulation combined with cycling or gait training was widely used to facilitate gait performance by relieving the subjects from “drop foot” (11, 49–51). It induced coordinated function assistance during ankle dorsi-flexion and reduced the energy consumption for stroke survivors during ambulation. Recently, Rouse revealed that muscle activation is improved in an FES cycling study; however, the performance of the clinical function was not evaluated (52). Previous studies reported that the absolute RMS value of paretic TA during maximum dorsi-flexion increased after FES intervention, which revealed that the capacity of the muscle output increased after training (53). In our study, the decrease of the normalized RMS value of TA after training might be attributed to the increase in maximum activation during MVC or less activation during pedaling, which could both reflect development in the muscle activation capacity of TA (54). In addition, the decrease of normalized RMS after training also suggested reduced excessive muscular activities of TA during pedaling tasks (55, 56). Furthermore, the reduction of excessive muscular activation indicated the development in voluntary motor controls (56). Yan et al. found a significant decrease of co-contraction ratio during dorsiflexion of the paretic ankle joint after FES intervention (without cycling) (51). It was explained and cross-validated that frequently repeated movements, induced by FES in this study, might reinforce network connection patterns. The FES-assisted cycling in our study allowed for continuous repetition of movements during pedaling. The assists from FES simulated and synchronized with normal muscle activation pattern from healthy subjects during cycling. Therefore, this synchronized stimulation and repetitive muscle contraction reminded the subjects how to pedal voluntarily and coordinately. An fMRI study reported that the repetitive FES-induced movements facilitated motor learning according to cortical mechanisms and spinal mechanisms (57), which revealed a more independent control of muscle contraction after training. Our study also demonstrated significant muscle co-contraction reduction within ankle movements during cycling (Figure 4). In addition, as shown in Figure 6, the sEMG enveloped signals of TA and MG were more rhythmic after training, the overlapping area of muscle activation decreased obviously. In Brunnstrom's theory, the reduction of TA/MG co-contraction could induce the separation movements of ankle joint and result in function improvement (58). This also explained our clinical results that the FMALE score increased, and the ambulation performance improved after training.

Our study combined EIM and sEMG techniques to jointly evaluate the FES-assisted cycling training effect. The increase of θ and AR value of TA after training may partly reflect that muscle cross-sectional area and muscle mass increased and might be related to the decrease in the intramuscular fat tissue, and the myofiber is arranged more regularly after training (17, 24, 25). Similarly, the current study proved that muscle activation from sEMG and muscle structure changes from EIM were both improved after the FES-cycling training. These results revealed the alteration of muscle properties after training from different points of view, which help us understand the training effect from different aspects.

Results of clinical scales indicated improvements in motor functions of lower extremity, gait, and balance after 20 sessions of training (Table 1). The walking distance in 6MWT improved 13.8% and the FMALE score increased 11.2% after training compared to that before training, which was similar to previous studies (59). In the study by Janssen, it was found that after FES-induced cycling training, the BBS score increased from 40 to 44.2 (10%) and 6MWT increased from 160 to 185 (15.7%) (11). However, the improvement of balance function after training was relevantly lower than that of the ambulation performance (6MWT) and motor function (FMALE) in the current study. Because most of the recruited subjects presented high BBS scores before training as shown in Table 1, the ability of balance control was not mainly required during cycling exercise (60). Thus, the subjects in this study presented limited improvement in balance function.

Another interesting result was that all the parameters of EIM and sEMG revealed that TA muscle presented the most significant improvement after training, and it was noticed that FMA scores of ankle segment presented a significant increase while no significant alteration was revealed by the FMA scores of knee segment. This outcome might be due to the higher stimulation intensity applied to the TA muscle as compared to the rest of the muscle groups according to our experiment records (Supplementary Material). This can be reasonable because the reduced innervation of TA was the leading cause of drop foot (61). Our correlation analysis also partly explained this unique improvement on TA. As shown in Figure 7, EIM, sEMG, and clinical scores significantly correlated with each other on TA. Therefore, our results partly reveal that after the training, muscle composition, and structure property of TA change toward the direction of nonparetic side of TA, and myofibers might be arranged more regularly and extra tissues (fat and connective tissues) are reduced (partly indicated by EIM), which might facilitate the increase of muscle activation volume during ankle dorsi-flexion (revealed by sEMG). Those alterations in some way possibly promoted ankle joint motor function and enhanced the walking performance. Based on the results, we may speculate that after the repetitive muscle contraction during FES-assisted cycling, the subjects might learn to coordinately control muscle activation (revealed by CI of sEMG), and this motor relearning procedure allowed subjects to perform better in ambulation and balance assessment.