Background
There are approximately 285,000 people in the United States living with spinal cord injuries (SCI) [
1], roughly two thirds of whom are dependent on scooters or wheelchairs for daily mobility [
2]. According to a survey of individuals with SCI, trunk stability is among the top functions they desire to improve or restore [
3]. People with SCI who have lost the ability to control their trunks experience more postural sway during sitting than able-bodied individuals [
4]. Minor disturbances can destabilize upright sitting and cause a loss of erect sitting posture during wheelchair use, leading to injurious falls and thus limiting their mobility and independence. Tips and falls are the leading cause of injury for people in wheelchairs, accounting for two-thirds of the more than 100,000 wheelchair related injuries per year requiring treatment in emergency departments [
5]. Wheelchair accidents can result in fractures, lacerations, contusions and abrasions, and even death [
6]. Rehabilitation for a fracture in an individual with SCI requires four to eight weeks of hospital stay, resulting in reduced strength and increased risk of blood clots [
6]. Additionally, cost of treatment due to wheelchair-related falls typically ranges from $25,000 to $75,000 [
6]. The riding surface and environmental factors [
7], wheelchair design (such as caster size and seat position) [
8], and center of gravity location [
9] can all affect wheelchair stability, and ultimately contribute to falls and tips experienced by manual wheelchair users. Common events that can contribute to destabilization include colliding with an obstacle (such as a low curb or wall), propelling through sharp turns, negotiating uneven or inclined surfaces, and maneuvering curb drops [
7,
10].
Current methods to maintain trunk stability for manual wheelchair users with SCI include seat belts to keep the trunk from falling forward, seat cushioning systems to decrease the exaggerated posterior pelvic tilt that results from paralysis of the hips and trunk, increased seat dump to take advantage of gravity in a backwards leaning posture, or supports to limit lateral motions of the trunk. These methods have many disadvantages, including reduced work volumes and impaired ability to reach and manipulate objects due to the external constraints, as well as pressure ulcers, skin tears, lowered self-esteem and even asphyxiation [
11]. Non-compliance with such strategies is high since they also restrict desired motions and interfere with functional tasks while the wheelchair is not moving [
12]. Power wheelchairs offer more options for specialized seating systems and can provide power tilting and reclining to allow users to independently change their postures in preparation for a task or anticipated disturbance. However, they impose the same restrictions on voluntary motion, and people are typically reluctant to switch from manual to powered wheelchairs for the additional stability as it may be viewed as an indication of greater disability [
13].
Functional neuromuscular stimulation (FNS) has been utilized to restore and maintain trunk stability by significantly increasing the trunk extension moment [
14] and multidirectional trunk stiffness [
15,
16]. When the trunk flexor and extensor muscles are activated with neural stimulation, people show physical and psychological improvements of strength and stability, particularly in response to unexpected, destabilizing forces [
17,
18]. EMG of healthy individuals has shown significant activation of the erector spinae muscles in response to perturbations pulling the trunk anteriorly [
19]. Continuous activation of the otherwise paralyzed hip and trunk muscles with FNS can maintain trunk stability during similar anteriorly-directed forces applied to the trunks of individuals with SCI [
17,
20]. Furthermore, activating the hip and trunk muscles with neural stimulation can restore upright sitting from forward flexed or side-leaning positions automatically without use of the arms based on exceeding an angular threshold monitored by a tilt sensor attached to the sternum [
21]. Similarly, a powered-wheelchair neuroprosthesis that activated trunk flexors and extensors with FNS on able-bodied participants experiencing perturbations in the anterior-posterior (AP) direction showed a decrease in trunk displacement and velocity with stimulation [
18]. Additionally, continuous electrical stimulation of the trunk and hip extensors has been shown to improve the mechanics of manual wheelchair propulsion for people with SCI by decreasing the peak resultant pushrim forces and increasing propulsion efficiency [
22]. These studies suggest that simple, automatic systems based on body position have the potential to improve mobility and voluntary function while sitting in a wheelchair, however automatically modulating stimulation at the onset of destabilizing events during manual wheelchair propulsion to minimize the effects of unanticipated perturbations or restore seated posture with neural stimulation has not yet been examined in person with SCI under simulated real-world conditions.
To employ neural stimulation in response to potentially destabilizing events, a method of accurately predicting or detecting such situations must be identified. Inertial measurement units (IMUs) have been utilized to monitor trunk angle of people with SCI in a static environment [
21], determine phases of the manual wheelchair stroke cycle [
23], and classify physical activities of daily living [
24]. Additionally, such classification of up to 90% of the activities of daily living of manual wheelchair users with SCI was found to be clinically useful [
24‐
26]. Prior investigation of classifying instability of manual wheelchairs combined use of IMUs and machine learning to differentiate destabilizing conditions [
27]. In contrast, our research utilizes a simple threshold-based method to detect and classify destabilizing events based on characteristics of the inertial signature of the wheelchair to automatically trigger activation of the appropriate muscles and determine its impact on seated posture.
The static and dynamic tipping stability of the wheelchair-user system has been examined in modeling studies [
28‐
31]. Li et al. [
30] derived an equation to calculate the critical velocity at which unrecoverable forward tipping will occur due to a collision. The equation and parameters used for this paper are explained in the methods section below. Similarly, Bruno’s collision simulations [
28] released a powered wheelchair from the top of a ramp to passively ride down into a barrier at the end, and found that the system is expected to fully tip forward when entering a collision at 1.74 m/s. Cooper et al. [
31] developed the following equation to predict rollover during a destabilizing turn, based on the initial velocity and radius of the turn. This equation is also explained in the methods section below. Such wheelchair modeling studies inform the design of wheelchairs for safe operation and define their inherent mechanical limits of stability, but do not react to potentially destabilizing events or assist wheelchair users in maintaining and/or regaining stable seated postures when they occur.
The purpose of this study was to determine the feasibility and preliminary performance of a threshold-based system that utilizes inertial measurements to detect destabilizing collisions and sharp turns in manual wheelchair users with SCI and provides an appropriate stimulated response of the hip and trunk muscles through implanted pulse generators (IPGs) or surface stimulation to enhance recovery of a stable seated posture, thereby potentially decreasing the risk of injurious loss of balance and falls. IPGs or surface stimulators electrically excite the motor nerves with adjustable pulse amplitudes and durations to activate the otherwise paralyzed hip and trunk muscles and generate the desired movements. It has been shown that FNS requires roughly 100 ms to generate torque in the muscles [
32]. Therefore, this system must act in real time to detect or predict the onset of such events to allow for earliest possible application of stimulation before instability reaches an irreversible state. We hypothesize that applying FNS at the onset of a destabilizing event will increase trunk stability, maintain erect posture, and improve users’ perceptions of safety during daily activities in their wheelchairs. This study analyzed detection accuracy, detection delay, and restoration of trunk stability with and without neural stimulation delivered by a threshold-based system during destabilizing collisions and turns in manual wheelchair users with SCI.
Discussion
A threshold-based system for maintaining trunk stability for manual wheelchair users was designed and tested in four subjects with SCI. The system designed and implemented in this study classified destabilizing collisions with an accuracy greater than 90%, which aligns with previous clinical applications of event detection devices [
24]. For S1 and S2, the stimulation significantly decreased the maximum AP trunk angle upon the collisions. Because of technical difficulties generating stimulation at the back extensor muscles in S3, FNS during these trials was limited to hip extensors. This limitation may have negatively affected the efficacy of the system to restore trunk stability for this subject. Despite attempts to control the speed at impact with the ramp and barrier set-up, the subject did not always achieve the required velocity of 1.5 m/s prior to collision to produce the desired deceleration profile which would have exceeded threshold and successfully triggered stimulation. A lower threshold or more precise control of the velocity at impact may have improved classification performance above the 93% accuracy observed.
During turns, the classifier also achieved detection accuracy greater than 90%. On average, delay in detection of right and left was 342 ± 73 ms, causing significant delay for stimulation to be activated. With an additional time delay of 100 ms for FNS to generate maximum torque [
32], the recruited muscles would not be activated fast enough to be provide the desired response in this application. The participants noticed this delay, stating their trunks felt more stable once stimulation was applied, but the stimulation became effective too late into the turn. One false negative was experienced for S3 during right turns. The thresholds for turns could be less conservative, allowing higher detection accuracy and possibly decreasing the delay of detection. The maximum ML trunk angle was not significantly decreased for any of the participants, and in some cases, increased when stimulation was applied. This may be a result of participants preparing for the sharp turns by using head and shoulder movements to lean into the turn and prevent falling out during trials with and without stimulation applied. To overcome the delay, future studies should consider monitoring the angular acceleration, rather than the angular velocity. Upon post-processing of the data across all four subjects, turn detection using angular acceleration achieved detection within 17.17 ± 11.40 ms. Alternatively, it may be necessary to activate stimulation before the event to effectively restore trunk stability. Investigation in other sensor modalities, such as ladar [
50‐
53] and vision-based methods [
53,
54], or sonar readings [
55] used to classify terrain in autonomous vehicle and robotic navigation applications, would be necessary to predict destabilizing events before they occur.
During a false negative classification, the controller does not apply FNS during a destabilizing event. In this case, if no secondary methods to maintain stability are in place, the user can fall out of the wheelchair and sustain injuries. Alternatively, a false positive classification would activate FNS during a period of stability. This activation of FNS may surprise the user, but will not produce any injurious consequences. Thus, in the event of a misclassification of this event detection controller, we would prefer false positive classifications.
It should be noted that because these trials were conducted in a laboratory setting and the events were controlled, some subjects assumed safety regardless of whether FNS came on or not as they understood the experiment would be designed to avoid potential injury. This assumption was reflected in participants’ URS scores, particularly S3, who rated all trials with and without FNS as moderately or very stable. Also, in this setting, we only tested the collision and turning events in the ways we designed them, which may not reflect how they occur in real life situations. For example, if an oblique collision happened, deceleration of this event would be detected in both the ML and AP directions, and may not be recognized as a collision using the algorithm discussed in this paper. The accelerometers utilized in this experiment were limited to a range of ±4 g. With this restriction, the mean and standard deviations used to define thresholds may be saturated as the AP acceleration often reached these limits during collision testing, though it is expected for 4 g to sufficiently define a destabilizing collision.
Previous work by Audu et al. [
17] has shown success using feedback controlled neuroprostheses for restoring an erect posture in people with SCI in response to external perturbations. Additionally, Patel et al. [
18] showed that a powered-wheelchair neuroprosthesis which activated FNS of trunk muscles of able-bodied participants in response to perturbations in the AP direction decreased trunk displacement and velocity. The work presented in this paper shows promise for an event detection system that could decrease wheelchair-related injuries, while increasing users’ independence without the need for restrictive straps or seatbelts. To our knowledge, this is the first study that applies neural stimulation in response to potentially destabilizing wheelchair events detected by wheelchair-mounted IMUs in individuals with SCI during simulated crashes and sharp turns. This system was tested with both implanted and surface stimulation, suggesting its potential applicability for a wider population beyond implant recipients. Larger scale studies with significantly mores subjects are necessary to determine the generalizability of the results and transcend the single-subject feasibility research design.
Future tests will be conducted utilizing accelerometers with higher ranges to fully understand the nature of the signals during collisions. This controller will also be tested outside of the laboratory setting for use in the community and during activities of daily living to understand the clinical significance of this device. Additionally, future tests will examine both a turn and collision simultaneously to ensure the controller responds appropriately to restore trunk stability during the combination of events. Lastly, algorithms will be developed for other potentially destabilizing events, such as rough terrain, bumps, and ramps. With increased trunk stability, manual wheelchair users may feel more confident traversing through unfamiliar environments and care should be exercised to capture such subjective perceptions of safety in addition to biomechanical measures.