Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Device Technologies and Biomedical Robotics

Device Technologies and Biomedical Robotics - Poster Session E

Poster J12 - Validation of candidate sensors for the design of a closed-loop system for somatosensory feedback restoration

Saturday, October 26, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

AF

Arnold Fleishman

Undergraduate Researcher
Oberlin College
Marina, California, United States

Primary Investigator(s)

LF

Lee Fisher, PhD

Primary Investigator
University of Pittsburgh/Rehab Neural Engineering Labs, Pennsylvania, United States

Co-Author(s)

GA

Gilgal Ansah

Graduate Student Researcher
University of Pittsburgh, United States
Dulce Mariscal, PhD

Postdoctoral Scholar
University of Pittsburgh, Pennsylvania, United States

Introduction: Restoring somatosensory feedback to individuals with lower-limb amputations has been shown to improve asymmetric gait, reduce fall risk, and alleviate phantom limb pain. Recently, our team has used lateral spinal cord stimulation (SCS), a minimally invasive procedure, to provide somatosensory feedback for individuals with limb loss. We implemented a closed-loop system, where the stimulation location and intensity are modulated by real-time input from a pressure sensor under the sole of the prosthetic foot. During gait, force applied to the pressure sensor is transduced into a signal that initiates electrical stimulation of the spinal cord. This stimulation evokes sensations that appear to emanate from the prosthetic limb. Here, we investigate the usability of candidate pressure-sensing devices that can be integrated into sensory feedback systems at a low cost. We ultimately aim to develop a wearable, affordable pressure-sensing insole that can be used in the daily lives of people with lower-limb amputations.

Materials and

Methods: We evaluated the force-to-voltage relationship and performance of two low-cost, commercially available force sensors: the Tekscan A502 force-sensing resistor (FSR) and 450N SingleTact capacitive sensor. The Tekscan FSR has a modifiable force sensing range capable of measuring up to 44,000N, while the SingleTact sensor has a force sensing range of 450N with some capacity for overload. For each trial, the sensors were loaded with a constant force for 60 seconds. The output voltage was fed to an inverting operational amplifier circuit and recorded with the Arduino Uno R3 microcontroller and corresponding Arduino Integrated Development Environment. Trials were repeated using different forces of up to 600N.
To quantify sensor performance, we measured the variability in the recorded voltage for each trial. We used the standard deviation of the voltage over the course of each 60-second trial as a metric of this variability. Trials with higher variability were classified as noisy signals and filtered using a 4th order butterworth low-pass filter. To characterize the force-to-voltage relationship of the sensors, we performed a regression analysis and measured the root mean square error (RMSE) to determine how well our model fit the data. To account for differences in voltage output range between sensors, RMSE was normalized to each sensor’s maximum output range.

Results, Conclusions, and Discussions: We found that the Tekscan A502 FSR output was significantly noisier than the SingleTact sensor output, with average standard deviations of 0.49V and 0.006V throughout each trial, respectively. To eliminate some of the noise in the Tekscan FSR, we implemented a 4th order low-pass butterworth filter. The filter greatly reduced noise, resulting in a signal with an average standard deviation of 0.082V, while having no significant effect on the median output throughout each trial (t(27) = 1.75, p = 0.091).
A linear fit effectively represented the force-to-voltage relationship for the Tekscan A502 FSR model (R2 = 0.966; p < 0.001; V = 0.00414 F + 0.143; see figure 1). On the other hand, a logarithmic model better represented the force-to-voltage relationship for the SingleTact sensor (R2 = 0.868; p < 0.001; V = 0.295 ln (F) - 0.442; see figure 2) than a linear model (R2 = 0.810; p < 0.001; V = 0.00130 F + 0.760). The model derived from the Tekscan FSR’s data achieved a more robust fit than the model derived from the SingleTact sensor’s data, with R2 values of 0.966 and 0.868, respectively.

In this study, we establish and describe a force-to-voltage relationship in two low-cost, commercially available sensors. The presence of this relationship opens the possibility of integrating these sensors into a closed-loop spinal cord stimulation system, where the voltage across the sensor regulates the stimulation intensity to deliver biomimetic somatosensory feedback. For this purpose, we recommend using the Tekscan A502 FSR because 1) the force-to-voltage relationship is more accurately predicted by the regression model, and 2) it has a higher force input range. However, further work should be aimed at optimizing filter parameters and determining the impacts of frequent loading cycles (that mimic natural gait) on sensor performance.

We conclude that there is a clear force-to-voltage relationship in both the Tekscan A502 FSR and the 450N SingleTact sensor. As a result, both of these sensors appear to be suitable for integration into a closed loop force-sensing system, with the Tekscan A502 FSR demonstrating superior performance.

Acknowledgements (Optional):