Functional electrical stimulation support
Based on previous work, we decided to use FES-induced flutter kicks for proficient front crawl swimmers. Furthermore, floats are attached to the ankles that lead to knee flexion and an upward movement of the ankle in a non-stimulated leg. On the one hand, this results in a more streamlined posture in the water. On the other hand, it implies that the desired knee movement can be realized by alternating between FES-induced knee extension and passive knee flexion caused by the floats. Hence, only two stimulation channels are needed. The quadriceps muscles of both legs are alternately stimulated where the stimulation electrodes were placed at the proximal part of the rectus femoris and the motor point of the vastus medialis of each leg. The stimulation, which is applied with stimulation pulse frequency of 25 Hz, is switched on and off at a rate of 1 or 2 Hz which results in approximately one or two leg kicks per arm stroke depending on the arm stroke frequency. The amplitude and pulsewidth can be varied in the ranges 0–100 mA and 0–500 μs, respectively. Both values are increased/decreased simultaneously to control the generated muscle contraction.
Transcutaneous spinal cord stimulation
Transcutaneous spinal cord stimulation is used with the aim to reduce lower-limb spasticity during and after swimming. Therefore, we stimulate the afferent fibers of the L2–S2 posterior roots continuously at 50 Hz using biphasic pulses with 1 ms pulse width over the T11/12 region at the spinal cord according to [8]. The electrode position at the back and stimulation amplitude has been determined as outlined in [8]. By switching on the tSCS, the trunk musculature is activated at a motor level as a positive side effect. This improves trunk stability and straightens the upper body. As shown in Fig. 1, a streamlined swimming position can be achieved with FES and tSCS compared to no stimulation in a paraplegic subject.
Additional file 1: Subject A.
Experimental setup
Stimulator
The stimulation system for swimming shown in Fig. 2 employs a CE-certified stimulator (RehaMove3, Hasomed GmbH, Germany) with customized firmware. A single current source is integrated into the device, and the output of the source is demultiplexed for up to 4 channels. The stimulator is placed inside a waterproof bag under the swimmer’s T-shirt. All stimulation cables are tunneled through the bag and drained with silicone to prevent water intrusion. The bag is attached with a strap on the swimmer’s back between the shoulder blades.
The stimulator can be controlled via the membrane keypad e.g. the stimulation program can be selected, started/stopped and the stimulation intensity can be adjusted. The stimulator is battery-powered, and the high-voltage source is galvanically isolated from the battery power. Hence, the current conduction is always constrained between the positive and the negative electrode of each stimulation channel.
Waterproof stimulation electrodes
Due to the fact that chlorinated water in swimming pools has a conductance of 2.5–3mS/cm, which results in resistance of 333–400 Ohm, a direct stimulation with non-waterproof electrodes would produce a parasitic short circuit between electrodes during stimulation. Therefore, the device-integrated electrode error detection might not detect a bad connection between the electrode and the skin. If both electrodes float in water, then the muscles would not be stimulated, because the current always takes the path of least resistance directly through the water and not the body. If only one electrode floats in water, then the current will still pass through the remaining firmly attached electrode and will still cause a muscle contraction beneath this electrode. The only potentially dangerous situation would occur when the conductive side of a detached and floating electrode would accidentally be firmly pressed against skin of the upper body, since then electrical currents might flow through sensitive organs, such as the heart. To minimize this risk and because of the limited electrode error detection, the electrodes need to be safely and firmly attached to the skin. Furthermore, the electrode side facing away from the body needs to be isolated against water. Possible measures are waterproof transparent film dressing, straps or swimming cloths.
Currently, there are no waterproof stimulation electrodes available on the market. Most transcutaneous electrodes consist of a conductive hydrogel adhesive which is connected via conductive film to a lead wire or metal snap stud and isolated with an insulative cover. If the hydrogel adhesive gets into contact with water it starts to absorb water while the thickness increases. Hence, the area with direct contact to the water increases. Furthermore, the adhesive function of the electrode is reduced. Approaches for underwater EMG measurement in [28, 29] used several layers of waterproof wound plaster with tunneled holes for the lead wires to waterproof standard adhesive EMG electrodes. The same procedure can be used for stimulation electrodes where standard electrodes are waterproofed with adhesive films, like TegadermTM or OpSiteTM.
For the training sessions of our pilot study, which is described in the next subsection, special electrodes developed by Axelgaard Manufacturing Co. Ltd have been used, as shown in Fig. 3a. A single electrode consists of a standard electrode with an oversized waterproof backing. The snap adapter is tunneled through this backing. The remaining task is then to connect the electrode lead (converter from the snap adapter to 2 mm socket) and seal it with a waterproof transparent film dressing (3M Tegaderm, 3M Co., USA). All cables and cable connections have to be waterproof as well. Otherwise, parasitic short circuits occur. Removable tight silicone tubes showed to be efficient in covering the connection between the electrode lead and the stimulation cable.
A drawback of adhesive electrodes with oversized waterproof backing is that after a single contact with water they cannot be reused. Hence, for each swimming session, a new set of electrodes is needed. To reduce costs and to save the environment, the suitability of reusable safety silicone electrodes shown in Fig. 3b to d has been investigated in a post-training assessment session. These electrodes are available in different sizes (VITAtronic Limited, Germany) and can be directly connected via a standard 2 mm electrode connector to the simulation cable. Due to the non-conducting upper side and the framed isolation on the conductive skin side, no parasitic short circuit can occur when firmly attaching the electrodes to the skin. The material is non-adhesive, which reduces skin irritation during the doffing phase but implies that it must be fixed with tight sleeves, straps, waterproof transparent film dressing, or with tight knee-length swimsuits. During swimming a small water film between the skin and the conductive part of the silicone electrode is present. Hence, no additional hydrogel was added. Straps and knee-length swimsuits have been used in this study for the leg electrodes. The electrodes for tSCS have been fixated by waterproof transparent film dressing.
Subjects, training protocol and outcome measures
This feasibility study was carried out at the Treatment Centre for Spinal Cord Injuries in BerlinFootnote 1. The aim of the study was to investigate the effects of stimulation-supported swimming in two SCI patients with complete paralysis of the lower extremities after spinal trauma with a lesion above Th10. Participants have to be proficient front crawl swimmers.
Both recruited subjects (A: age 40, time since injury 10 years, B: age 58, time since injury 36 years) are ASIA impairment scale A with lesion level Th5/6 and gave written informed consent. They both complain of a moderate clonus of the lower extremities and the abdomen during position changes, and Subject A experiences leg extensor spasms from time to time. Subject B suffers from a hip joint contracture.
After the recruitment and initial assessment, the subjects were asked to carry out a four-week FES cycling training at home. During this land training, they trained at least three times a week for 30 min with a standard FES cycling ergometer (RehaMove, Hasomed GmbH, Germany). This preliminary FES cycling training was needed to build up a defined baseline strength and endurance for the swimming phase. During the swimming phase, FES cycling activity was reduced to two times a week.
The entire swim training lasted for 10 weeks. Subjects were asked to attend the weakly swim training session that lasted between 30 to 45 min (excluding donning and doffing). As a safety measure, the swim sessions were always accompanied by a trained pool guard. Furthermore, all recruited subjects are able to swim without stimulation. The training was done at a 16 m pool. Subject A used a snorkel during front crawl swimming.
Prior to the first use of tSCS during swimming, the electrode position at the spinal cord and the stimulation intensity for spasticity treatment were identified according to [8] and documented. The found constant stimulation intensity was applied in all training sessions when tSCS was on.
The stimulation amplitudes for both quadriceps were identical and have been chosen to cause an almost full knee extension while the subjects rested at the edge of the swimming pool with an upright upper body. Before each lap, the leg movement was reevaluated and the stimulation amplitude increased, if necessary, to compensate for muscle fatigue. A break of at least one minute was kept between the laps.
At the beginning of each swim training session, lap times were measured. Therefore, the subjects were instructed to swim each 16 m lap as fast as possible. When comparative measurements were taken, first the times for swimming without support were taken, then with FES support and finally the times for FES plus tSCS support. We used this order so that the results for trials with increasing amount of support are more affected by muscular fatigue then the trials with less or no support. After this initial assessment, training with the preferred support (FES or FES plus tSCS) took place for the rest of the session at self-selected swimming speed. If FES plus tSCS has been selected as preferred support, then tSCS was always active also in the breaks between the laps, while FES was switched off during these breaks.
There are three main questions that shall be answered in this pilot study:
Does the swimming speed, assessed by lap times, increase compared to non-assisted swimming?
Does the general well-being of the subject improve during the trial?
How is the acceptance of the technology by the user?
The subjects were asked to rate the therapy on the basis of predefined statements using a five-grade scale between full agreement and no agreement. Using the result of the questionnaire the last two questions can be answered.
IMU-based motion analysis during swimming
Post-training assessment
Nine months after completion of the entire swim training phase, after we had acquired a suitable measurement system, we performed an additional swimming session with each of the two subjects to monitor the effects of the different stimulation programs on the leg and trunk motion. Both subjects were instructed to repetitively swim laps with no support, tSCS support, FES support, and FES plus tSCS support as fast as possible.
Sensor setup
A wearable sensor setup was used. The employed system WaveTrack (Cometa srl, Italy) is a wireless and waterproof inertial sensor system consisting of several time-synchronized inertial measurement units (IMUs). These inertial sensors provide three-dimensional measurements of the acceleration, angular velocity, and magnetic field vector at a frequency of 286 Hz. The sensor data were used to determine the joint angles of both knees and both hips as well as the roll orientation angles of the trunk on the cervical and lumbar level. To this end, four IMUs were bilaterally attached to the exterior thigh and shank, and two IMUs were located on the upper and lower back, as shown in Fig. 4a and b. Note that only the left leg is depicted. For both IMUs on the right leg, the local x-axis points longitudinally toward the feet, but the z-axis points laterally to the right, which implies that the y-axis points anteriorly.
As all of the sensors are located underwater during the whole measurement, wireless data transfer (streaming) is not an option. Therefore, an offline data recording is carried out. The data acquisition and time synchronization of the sensors is initiated by means of remote control. The recording begins before the subject enters the pool. After leaving the pool the recording is stopped and the data are transferred from the sensors to a PC. The software EMGandMotionTools (Cometa srl, Italy) was used for data transfer and sensor settings. Admittedly, due to the loss of communication between the sensors when located underwater, a synchronization drift is educed. However, since this drift does not exceed a few milliseconds per hour and all acquisitions last between approximately 30 to 45 min, the effect on the data is considered irrelevant.
All sensors were attached to the skin by means of double-sided adhesive tape for rough fixation. Subsequently, a transparent 3M Tegaderm film was used in order to prevent movement and loosening of the sensors during the swimming process.
Joint and roll angle estimation
For each body segment, the IMU readings are used to estimate the segment orientation with respect to an inertial frame of reference. To avoid the assumption of a homogeneous magnetic field inside the building and especially inside the water, we refrain from using the magnetic field vector measurements and fuse only the measured accelerations and angular rates by using a modular quaternion-based sensor fusion algorithm [32]. It must be noted that orientations obtained by such a 6-axis sensor fusion cannot be used for joint angle calculation directly since they exhibit an arbitrary heading offset and drift slowly around the vertical axis. With accurate bias estimation, that drift can be as slow as one degree in ten seconds, but it will not be reduced to perfect zero.
To overcome this drawback of the magnetometer-free approach, we exploit approximate kinematic constraints of the hip and knee joints. During the considered flutter kick motion of the legs, the hip and knee move approximately like hinge joints – flexion/extension is the dominant motion, while adduction/abduction and internal rotation occur only to a limited degree. We exploit these approximate kinematic constraints by using a recently developed relative-heading tracking algorithm [33]. That algorithm takes the orientation quaternions of both segments adjacent to the joint and corrects the heading of the distal segment’s orientation such that the joint constraint is fulfilled in a weighted least-squares sense. We apply this method repeatedly, starting from the lower-back segment and moving distally towards the shanks.
Consequently, we obtain seven quaternions that describe the body segment orientations with respect to a common inertial frame of reference. We can thus calculate joint angles from these quaternions. The relative joint orientations are found by multiplying the conjugate of the proximal orientation with the distal orientation. The joint angles are then calculated by intrinsic Euler angle decomposition of this relative orientation quaternion. Note that both the hip and knee extension angles are defined such that they are 180 degrees for a perfectly straight leg.
Finally, the roll angle of the upper and lower back is determined from the corresponding orientation quaternion. This is achieved by transforming the local left-to-right axis, i.e. the y-axis of the IMU, into the inertial frame of reference and then determining the angle between that axis and the horizontal plane, as illustrated in Fig. 5. Note that this angle is defined positive when the right side of the trunk is lower than the left side.
A segmentation of the recorded data is performed based on the norm of the 3D acceleration vector by detecting rest and motion phases. Only the first lap of each support modality is exported and investigated. From the extracted lap data, a time course over 7 strokes in the middle of the lap has been selected to analyze the joint and roll angles by using boxplots. Consequently, the start and stop phases of each lap are excluded from data analysis.