Mechanisms of neuro-robotic prosthesis operation in leg amputees

Abstract

Above-knee amputees suffer the lack of sensory information, even while using most advanced prostheses. Restoring intraneural sensory feedback results in functional and cognitive benefits. It is unknown how this artificial feedback, restored through a neuro-robotic leg, influences users' sensorimotor strategies and its implications for future wearable robotics. To unveil these mechanisms, we measured gait markers of a sensorized neuroprosthesis in two leg amputees during motor tasks of different difficulty. Novel sensorimotor strategies were intuitively promoted, allowing for a higher walking speed in both tasks. We objectively quantified the augmented prosthesis' confidence and observed the reshaping of the legs' kinematics toward a more physiological gait. In a possible scenario of a leg amputee driving a conventional car, we showed a finer pressure estimation from the prosthesis. Users exploited different features of the neural stimulation during tasks, suggesting that a simple prosthesis sensorization could be effective for future neuro-robotic prostheses.

Fig. 1. Nerves-connected leg prostheses enhance mobility during OT and ST.

(A). Schematic representation of the sensorized prosthetic leg and its main components. The involved subjects were transfemoral amputees equipped with (i) neural interfaces in their tibial nerve, (ii) RHEO KNEE XC with an angle knee encoder, (iii) sensorized insoles, (iv) microprocessor programmed for a real-time acquisition and conversion of the sensors’ readouts to neural stimulation, and (v) a portable neurostimulator. Real-time tactile sensory feedback was provided using three electrode channels eliciting sensations in three positions under the phantom foot associated with three sensors of the insoles. Real-time position feedback, related to the prosthetic knee flexion/extension, was also provided using an electrode channel evoking a sensation referred to the phantom calf muscle. (B) OT. A subject while walking over ground, synchronous sensorized insole and knee encoder readouts, and encoded currents injected into the TIMEs. (C) ST. A subject while climbing (left) and descending stairs (right). Synchronous sensorized insole and knee encoder readouts, and encoded currents injected into the TIMEs. In both tasks, markers were placed on the main leg joints (Materials and Methods) for the kinematics gait analysis. (D) Box plot of the gait speed (meter per second) with (SF) and without (NF) neural sensory feedback during OT for subjects 1 and 2. n = 30 sessions per condition. (E) Box plot of the gait speed (number of laps per session) with (SF) and without (NF) neural sensory feedback during ST for subjects 1 and 2. n = 24 sessions per condition. Box plots of the self-reported confidence (F) during OT and (G) ST for subjects 1 and 2 in NF and SF are displayed. P values of the Wilcoxon test are shown. (Photo credit: Giacomo Valle, ETH Zurich).

Fig. 2. Sensory feedback activates novel motor strategies.

(A and B) The length of the stride during OT is displayed for subject 1 (n = 45) and subject 2 (n = 50). Cadence during OT is reported for subject 1 (n = 100) and subject 2 (n = 110). (C) Stance time during OT is reported for both legs (healthy and prosthetic) in both subjects. n = 46 for subject 1 and n = 30 for subject 2 for condition and leg. (D and E) Stance time and swing time during the ascending phase of the ST are shown for both subjects and legs in NF and SF conditions. For the prosthetic leg, n = 24 for both subjects while for the healthy leg n = 22. All data are shown as box plots with (SF) and without (NF) neural sensory feedback during OT and ST. P values of the Wilcoxon test are shown. (Photo credit: Giacomo Valle, ETH Zurich).

Fig. 3. Quantification of the augmented prosthesis confidence.

(A) The vGRF was calculated on both legs using the pressure sensors of the sensorized insoles. The loading peaks and the push-off peaks were measured on both legs. (B) The loading peaks measured on the healthy side showed a difference between SF and NF in both participants. n = 48 per subject and per condition. (C) Loading peaks and push-off peaks measured during the ascending phase in ST on the prosthetic side in both subjects are shown. n = 24 per subject and per condition. (D) Push-off peaks measured on the healthy side during the ascending phase in ST in both subjects are displayed. n = 16 per subject and per condition. All data are shown as box plots for neural sensory feedback (SF) and for without (NF) during OT and ST. P values of the Wilcoxon test are shown. (Photo credit: Stanisa Raspopovic, ETH Zurich).

Fig. 4. Underlying code of neural stimulation.

Stimulation channels activated during OT (A), ST ascent (B), and ST descent (C). Percentage of steps with sensors activated during these tasks according to the steps (OT, n = 50; ST ascent, n = 70; ST descent, n = 50). In OT, the percentage of steps with sensors activated in order (first, heel; second, lateral; and third, central) is shown. On ST ascent, two cases occurred: three (central, lateral, and heel) or two active sensors (central and lateral). On ST descent, only two sensors were activated (lateral and heel). Different temporal order or spatial usage could be a simple, but robust, indicator of intuitively integrated codes for different motor behaviors. (Photo credit: Stanisa Raspopovic, ETH Zurich).

Fig. 5. Reshaping of kinematics toward a more physiological gait.

(A) During OT, main joints markers of the prosthetic leg and of the healthy leg were tracked over time from the sagittal plane. The trajectories, peaks occurrences, and velocities were calculated for each subject. (B) For both subjects, horizontal ankle velocities of the prosthetic leg were different in SF with respect to NF (n = 24 per condition). Double support as percentage of gait cycle is presented for both subjects in NF and in SF. Dashed line indicates reference normative value. n = 46 for subject 1 and n = 30 for subject 2 per condition. (C) During ST, the leg markers of the prosthetic leg were tracked. Ankle elevation during the initiation (first step) and steady state (second and third steps) are displayed for subject 1 (n = 24 per condition) and subject 2 (n = 25 per condition). (D) The peak occurrences of the ankle elevation and the velocities (x and y) are shown for each subject during the descending phase of the ST (only for the step with the prosthetic leg, n = 24 per condition). All data are shown as box plots for neural sensory feedback (SF) and for without (NF) during OT and ST. P values of the Wilcoxon test are shown. (Photo credit: Giacomo Valle, ETH Zurich).

Fig. 6. Proof-of-concept scenario for the future neuro-leg prostheses.

(A) During the fine pressure reproduction task (PRT), the participant was blindfolded and acoustically shielded. When the subject pressed on the pedal to reproduce the level of pressure requested, the simultaneous readout from insole sensors were feed back to the user through intraneural stimulation using TIMEs. The requested level was recorded in revolutions per minute (rpm). (B) rpm measured during PRT according to the level requested. Data are shown as box plots with neural sensory feedback (SF), with position feedback (P), and without neural sensory feedback (NF) (n = 48). In plots: 1, low; 2, medium; 3, high force levels. P values of the Kruskal-Wallis test with Tukey Kramer for multigroup correction are shown. (Photo credit: Giacomo Valle, ETH Zurich).