Neurorobotics for neurorehabilitation | Science

Shifting the Focus Toward the Lower Limbs

Until recently, most research has focused on hand amputees, neglecting the clinical reality that four out of five amputees have lower-limb loss. Subjects with lower-limb amputation frequently do not engage fully in everyday activities because they are afraid of falls and do not perceive the prosthesis as part of their body (low “embodiment”). Such individuals often report poor satisfaction with their prostheses, citing the prosthesis as an excessive weight, despite prosthetic limbs typically being less than half the weight of a natural limb (11). They also tend to have reduced mobility (12), which can induce a sedentary lifestyle that promotes disease development and hinders reinsertion into society. PLP is also common and is poorly managed with current medications (13). Additionally, those with lower-limb amputations face substantially higher metabolic costs while walking, resulting in an increased risk of heart attack compared with the general population (14).

We have pioneered a human-machine system that translates prosthetic sensors’ readouts into “language” understandable by the nervous system, using a detailed computational model (9, 10) that indicates an optimal number of implants for the targeted nerve. Compared with traditional frequency variation, the model favors current amplitude modulation, for increasing efficiency of mapping and closed-loop stimulation (15).

Placement of the cable connecting implants to the stimulator is a longstanding problem and a frequent cause of failure in implantable technologies during clinical testing (7, 8). We addressed this issue during surgical preparation by implementing a release loop and stabilization within the fascia tissue graft with cables embedded in the middle (14).

Regaining the Functionality

We developed a “sensing leg,” for lower-limb amputees, by connecting sensors from the prosthetic knee and under the foot to the residual PNS (see the figure). An effective connection was achieved by equipping a microprocessor-controlled prosthesis with a purposely developed sensorized insole. An external controller that communicates wirelessly with the “sensorized prosthesis” proportionally transduces the readout of the insole and knee sensors into stimulation parameters. The stimulator then injects the current into the intraneural electrodes, eliciting sensations from the missing lower limb. The whole process ran at a delay unperceivable to the user, enabling real-time neuromodulation dependent on leg status.

This intervention enabled recovery of rich leg and foot perceptions (such as touch, proprioception, and both simultaneously). Users were able to recognize when the prosthetic leg—physically disconnected from their body and communicating wirelessly with the implants—was touched over different foot positions, flexed, or both. Users were also able to avoid a substantially higher number of stumbles when walking over obstacles, while wearing glasses that blind their lower field of vision, than when not exploiting the restored feedback.

Climbing stairs is often a challenge for above-knee amputees, resulting in very slow motion and considerable fatigue. When neuromodulation restored limb perception, their mobility substantially increased (15).

After these laboratory tests, volunteers stepped outside into a more natural environment. Because of the fully portable neuromodulating system, their confidence was increased, and subjects were able to walk with increased speed over what would normally be challenging sandy terrain. At the same time, volunteers’ metabolic consumption was diminished when sensory neurofeedback was switched on. The decreased energy expenditure when using neuromodulation could potentially limit cardiovascular system fatigue—a tremendously important health benefit for lower-limb amputees (14).

Modulating the PNS to Address Central Issues

When the implantable system was used in “neuro-pacemaker modality”—stimulating the nerve without connection to the prosthesis—a reduction in PLP was observed. Through precise somatotopic stimulation, we have evoked pleasant, close-to-natural sensations within regions of referred pain. By contrast, commercial stimulation devices mainly deliver prefixed and often ineffective patterns of stimuli, which do not elicit physiologically plausible sensations and fail to deliver effective relief (16), and spinal cord stimulators involve the induction of paresthesia (an uncomfortable tingling), which does not always completely relieve the pain (17). Meanwhile, neurostimulators that directly target the peripheral nerve deliver either nonselective stimulation or induce an analgesic nerve block, both of which have considerable drawbacks (18). Our neuromodulation pain treatment represents a real advance with respect to existing treatments, in that we restore naturalistic percepts, revitalizing the physiological pathway for sensations. Beside the imminent pain relief, this potentially induces beneficial long-term neuroplastic changes at the central nervous system (CNS) level, offering not only an analgesic but also a “curative” effect.

As a consequence of the restoration of physiologically plausible sensations, subjects experienced (“embodied”) the prosthesis similar to a real limb. Embodiment is typically measured in “nonfunctional” scenarios [such as rubber-hand experiments (19)]. We were able to measure an objective functional embodiment (20) increase during our experiments with the bionic leg, with and without feedback (15). Increased neural embodiment decreased weight perception (11)—a subjective percept influenced by cognitive processes. Brain cognitive load, measured with electroencephalography, also decreased while walking with the neuroprosthesis and performing a dual task (15).

Unveiling Neurorobotics Mechanisms

Neuromodulation triggered by a robotic device influences sensorimotor strategies employed by users, by means of its integration into their “traditional” nervous system. To better understand underlying mechanisms, we measured gait features of leg amputees during motor tasks of different difficulty while using the neuroprosthesis. They performed an easy task (walking over ground) and a challenging task (ascending and descending stairs) while gait and neurostimulating parameters were collected. The neuroprosthesis reshaped subjects’ legs’ kinematics toward a more physiological gait owing to sensorimotor strategies that allowed users to intuitively exploit various features of the neural code during different tasks (21). These strategies included different temporal order, or spatial usage of stimulation channels, resulting in simple but robust intuitively integrated neural codes for different motor behaviors. In a hypothetical scenario, which required a leg amputee to simulate driving a conventional car, we demonstrated a finer pressure estimation from the prosthesis, suggesting that even a simple neural code could effectively improve wearable neuromodulating devices.

These studies not only provided clear evidence of the benefit of neuromodulation for lower-limb amputees but also provided insights into fundamental mechanisms of supraspinal integration of the restored sensory modalities. Even with only a limited restoration of sensations from foot and knee, the CNS was able to successfully integrate and exploit this information. Analogous findings were observed in animals that compensated for lack of a single sensory modality through supraspinal structures (22).

Toward Widespread Use

The health benefits achievable from neuromodulation are of paramount importance to millions of impaired individuals. Because the economic cost of such technologies remains considerable, it is important to emphasize the accompanying benefits, which could eliminate the need for treatments related to pain or cardiovascular problems (1).

Together with pioneering results in neuromodulating treatment for neuropathy, the described research presents a conceptually new framework for neuroprosthetic device design, implementation, and testing. This iterative framework consists of (i) developing a deep understanding of the problem through models and experiments, (ii) influencing the device design, and (iii) a meticulously planned clinical testing phase. Multifaceted validation of experiments—including functional, emotional, and cognitive outcomes—feeds back to increase our knowledge and further optimize design. Model-based, deep understanding of the effects of neuromodulation could benefit future projects in the emerging field of bioelectronic medicine (23, 24).