Overground vs. treadmill-based robotic gait training to improve seated balance in people with motor-complete spinal cord injury: a case report

We used an alternating treatment design with three intervention phases to compare the Ekso and Lokomat methods of robotic gait training. The two groups were Ekso-Lokomat-Ekso and Lokomat-Ekso-Lokomat, with 10 training sessions in each intervention phase for a total of 30 sessions and no washout between intervention phases. Randomized allocation of the participants to the groups was concealed at the examination. Seated balance control outcome measures were repeated at the end of each intervention phase.

Participants performed up to 45 min of robotic-assisted gait training 3–4 times per week. Rest breaks were provided when needed by the participant. Participants reported their rating of perceived exertion (RPE) on the Borg CR-10 Scale every 10 min during training [14].

Initial training with the Ekso™ (Ekso Bionics, California, USA) consisted of sit-to-stand, standing balance, weight shifting, and stand-to-sit functions. Training focused on improving walking performance (triggering steps, executing a turn, and stopping). All participants were in ‘ProStep’ mode, wherein steps are automatically triggered when the weight-shifting targets are achieved, within the first session. Auditory cues for the weight-shifting targets were provided as feedback, and removed when an efficient gait pattern was maintained. The 10-meter walk test (10MWT) was used to record their fastest possible gait speed during training, which will reflect their ability to weight shift and trigger steps efficiently. The best time of three trials was recorded. For each session, we documented ‘Up Time’ (combined standing and walking time), ‘Walk Time’ (walking time only), and the number of steps taken.

Gait training with the Lokomat robotic gait orthosis focused on increasing gait speed. Treadmill speed was set to the fastest speed that the patient could tolerate, and subsequently increased by increments of 0.1 km/h every 10 min. If spasticity was exaggerated or poor foot contact with the treadmill was observed, the speed was lowered by 0.1 km/h. The level of BWS was adjusted to the minimum tolerated by the patient while ensuring appropriate stance phase kinematics. For each session, we documented BWS, walking speed and total distance.

Two baseline assessments of seated balance control were conducted prior to training and separated by 1 week. The assessment was repeated within 1 week of the final training session at the end of each intervention phase. We evaluated seated balance control by asking participants to sit on a forceplate (Bertec; Columbus, Ohio) covered with a foam pad. The forceplate was elevated so that the feet were off the ground. Participants were instructed to sit as still as possible while performing two static sitting balance tasks (quiet sitting with eyes open and with eyes closed) for 60 s with their arms crossed at the chest (Fig. 1a). During the eyes open task, individuals focused their gaze on a target 10 ft away. This protocol has been previously used to verify impaired seated postural control in people with SCI [15]. We also performed a dynamic sitting task to evaluate their LOS in the eight cardinal directions. During the LOS test, each trial started with 20 s of quiet sitting with eyes open to establish a baseline limit, calculated as the COP mean position plus four times the standard deviation (SD). Then, we provided visual biofeedback of their COP position (e.g. green dot) and baseline limits (e.g. a red box scaled to the baseline limit and centered at the mean COP position) on a computer monitor. Participants were instructed to lean as far as they could without losing their balance and return their COP within the baseline limit (Fig. 1b). There was no time constraint to complete the movement and the order of each direction was randomized. The movement direction was presented as a yellow arrow on the monitor. Two trials were recorded for each static balance task and the LOS test. This assessment was repeated 1 week later to establish a stable baseline. Data were collected at 100 Hz and stored for offline analysis.

These data were subsequently used to calculate the root mean square distance (RDIST) and velocity (RVEL) from the mean COP position to examine overall postural stability and the amount of postural activity during the static balance tasks [16]. We also calculated a 95% confidence ellipse area (AREA-CE) during static balance to measure stability performance [16]. The LOS test was scored as the total distance (sum of the maximum distance between the COP mean position and the furthest point in each direction). A mean was calculated per condition for all outcome measures.

Participants also performed two clinical measures of seated balance function at baseline and after each intervention phase. The T-shirt Test measures the time taken by the participants to don and then doff a T-shirt. The mean of two trials was calculated with the total time. For the Modified Functional Reach Test (mFRT), participants sat unsupported with their hips, knees, and ankles at 90°, then pushed a ruler forward with both arms and held their maximum position for at least 2 s. The mean of three trials was calculated with the distance. These tests of unsupported sitting have been proven reliable and valid in SCI [17, 18].

When participants were able to walk overground with the Ekso in ‘ProStep’ mode at a gait speed of at least 1.0 km/h, which is the minimum speed of the Lokomat, we conducted an assessment of trunk muscle activity comparing each method of robotic-assisted gait. EMG data were recorded using surface electrodes (SX230-1000, Biometrics Ltd., Newport, UK) connected to a portable data acquisition system (DataLOG, Biometrics Ltd., Newport, UK). Electrodes were placed on the right side for the rectus abdominis (RA)–3 cm lateral and 2 cm caudal to umbilicus; external oblique (EO)–2 cm below the lowest point of the rib cage [6]; and erector spinae (ES)–2 cm lateral to the T3, T12 and L4 spinous processes [19]. EMG signals were recorded at 1000 Hz. Foot switches were used to determine heel strike and toe off for each step. Participants performed or attempted a maximum voluntary contraction (MVC) at the trunk for flexion, lateral flexion, rotation and extension. Baseline EMG activity for all muscles was recorded while participants were lying supine.

Participants performed two walking conditions at matched speeds: 1) Ekso and 2) Lokomat. Three trials were recorded per condition. Overground speed was calculated from the time taken to traverse the middle 10 m of the 12 m path. During Lokomat walking, we used a standardized BWS level set at 50% of the participant’s body weight.

The average EMG amplitude recorded during MVC was used for normalization. EMG data was filtered with a sixth-order dual pass Butterworth filter at a high-pass of 30 Hz to remove electrocardiography artifact, then rectified and filtered with a sixth order dual pass Butterworth filter at a low pass of 50 Hz using custom routines written in MATLAB (Mathworks, Natick, MA, USA). Data were separated into steps synchronized to right heel strike. The average time series amplitude of at least 20 steps was calculated per condition. Then, the root mean square (RMS) of the EMG signal over each gait cycle for each muscle was calculated to quantify the EMG amplitude of the trunk muscle activity. RMS was also calculated for the baseline EMG activity for all muscles during static supine position.

Baseline postural sway measures indicate that participants had greater instability during eyes closed compared to eyes open (Fig. 3a & b). During the first baseline assessment, static seated balance control with eyes closed was difficult for P1, who lost his balance and had to grab a handrail for recovery. Data on this trial was analyzed prior to the handrail recovery. All other trials were completed successfully.

After the first intervention phase of Ekso training, P1 and P3 reduced the RDIST, RVEL, and AREA-CE of the COP during both EO (Fig. 3a) and EC (Fig. 3b) conditions, indicating improvement in quiet sitting performance. After Lokomat training, there was a general increase in RDIST and AREA-CE, indicating worsening of performance. This was followed by general improvement in these measures (decreased scores) after the final Ekso training phase, except for the EC condition with P3. P1 continued to reduce RVEL across the Lokomat and final phase of Ekso training, while it increased back to baseline values for P3. The RDIST and AREA-CE remained similar for P2 between both baseline assessments and after the first intervention phase of Lokomat training in both EO and EC conditions. P2 reduced RDIST and AREA-CE after Ekso training compared to baseline and Lokomat training, with a larger effect in the eyes closed condition. RVEL showed no change across assessments for P2 with EO, but an increasing trend for EC with the exception of a slight decrease after Ekso training.

The change in LOS total distance showed a similar trend for P1 and P3: increase after Ekso training compared to baseline (indicating improved performance), then decrease after Lokomat training, and subsequent increase after the final Ekso training (Fig. 3c). P2 showed no change between baseline 2 and Lokomat training for LOS total distance, followed by an increase after Ekso training, and then a decrease after the second Lokomat training phase (Fig. 3c).

All participants slightly reduced their total time during the T-Shirt Test and increased their distance during the mFRT after Ekso training (Table 2). There was a tendency for greater time taken on the T-Shirt Test and shorter distance during the mFRT after Lokomat training as compared to Ekso training (Table 2).

The RA and EO muscles showed tonic activity over the gait cycle during Ekso and Lokomat walking, while the ES muscles showed a burst of activity at the transition from stance to swing that is more noticeable in the Ekso condition (Fig. 4a). Ekso walking produced higher trunk muscle activity compared to Lokomat walking in all participants (Fig. 4b). In fact, mean EMG amplitude in all muscles during Lokomat walking was similar to that recorded during baseline.