Pre and Post-Processing in Prosthetics

Prosthetics are incredibly important to many people who lose their limbs in order to regain a semblance of normalcy and independence. The field has come a long way from stiff limb extensions to fully functional prosthetics capable of performing most functions of regular flesh-and-blood limbs. However, while prosthetic limbs have come close, they're still not quite up to the level of flesh-and-blood limbs. That's where algorithms and processing power come in.

In the article "Intuitive Control of a Powered Prosthetic Leg During Ambulation," Dr. Hargrove and colleagues wanted to see if pattern recognition algorithms inside a powered prosthetic could utilize EMG signals to anticipate a user's intent for what specific movement they wanted to perform, and then allow and power that movement with little to no error. To do this, they got seven participants, most of whom had above-the-knee amputations. The participants were fitted with prosthetics with sensors inside that adjusted its mechanical response depending on the user's EMG inputs. Each participant went through a series of trials involving different types of walking (e.g., level-ground, ramp ascent and descent, stair ascent and descent) in order to train the pattern recognition algorithms inside of the prosthetic. The researchers tested different combinations of algorithms, including the linear discriminant analysis (LDA), which is able to average sensor snapshots, and the dynamic Bayesian network (DBN), which is able to interpret the anticipated trajectory of a desired movement based off the sensors. The researchers found that the participants preferred the mech + EMG + DBN combination, since errors were significantly reduced while walking. The prediction-based, pre-processing DBN algorithm seemed to help minimize errors while walking.

As computing and algorithms grow more sophisticated, post-processing may also be a viable way to help reduce prosthetic error during walking, as seen in this next study.

In the article "Probability-Based Rejection of Decoding Output Improves the Accuracy of Locomotion Detection During Gait", Dr. Ahkami and colleagues wanted to see if they could improve prosthetic walking by adding post-processing algorithms to the prosthetic, like LDA with rejection-based post-processing. To do this, they got 21 able-bodied participants, and used the LocoD dataset consisting of EMG, IMU, and pressure sensor data from the participants. Each participant had sensors on them to have their data collected, and went through a series of trials involving different types of walking, similar to the previous study mentioned, with different combinations of algorithms. The researchers then did some computational analysis to see which algorithms best detected the participants' movement within the data. The researchers found that rejection-based post-processing increased the accuracy of movement detection in the prosthetics' control algorithms.

Combining both pre-processing and post-processing algorithms into prosthetic limbs may be able to replicate the natural human motor movement in a prosthetic to the point where it feels like the user didn't lose a limb in the first place. While there's still work to be done, both of these studies show very promising findings for the field, and give hope to those with limb amputations for a chance at normalcy and independence through their prosthetic limbs.

References: 

Ahkami, B., Just, F., & Ortiz-Catalán, M. (2023). Probability-Based Rejection of Decoding Output Improves the Accuracy of Locomotion Detection During Gait. 023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). https://doi.org/10.1109/embc40787.2023.10340993

Hargrove, L. J., Young, A., Simon, A. M., Fey, N. P., Lipschutz, R. D., Finucane, S. B., Halsne, E. G., Ingraham, K. A., & Kuiken, T. A. (2015). Intuitive control of a powered prosthetic leg during ambulation. JAMA, 313(22), 2244–2252. https://doi.org/10.1001/jama.2015.4527