Intent Recognition for Prosthesis Control

Not Applicable

Completed

Conditions: Amputation

Interventions: Device: Robotic Knee/Ankle Prosthesis

Registration Number: NCT05537792

Lead Sponsor: Georgia Institute of Technology

Brief Summary: This work will focus on new algorithms for powered prostheses and testing these in human subject tests. Individuals with above knee amputation will walk with a robotic prosthesis and ambulate over terrain that simulates community ambulation. The investigators will compare the performance of the advanced algorithm with the robotic system that does not use an advanced algorithm.

Detailed Description: The focus of this work is a proposed novel AI system to self-adapt an intent recognition system in powered prostheses to aid deployment of intent recognition systems that personalize to individual patient gait. The investigators hypothesize that the prosthesis using our self-adaptive intent recognition system will improve walking speed. Independent community ambulation is known to be more challenging for individuals with TFA, and so the investigators will measure self-selected walking speed (SSWS) which is a correlate with overall health and is a predictor of functional dependence, mobility disability and falls; furthermore, slow SSWS are correlated to lower quality of life (QOL), decreased participation and symptoms of depression. Self-adapting intent recognition has great potential to restore gait in community settings and improve embodiment, which has been associated with improved QOL and increased device usage in patients who use advanced upper limb prostheses. In this experiment, patients with TFA will be fit with our robotic knee/ankle prosthesis and proceed to walk over a treadmill and overground at varying speeds, while the investigators capture 3D biomechanics in both the self-adaptive and static user-independent system (control condition). The investigators expect the self-adaptive system to learn the best prediction of the patient's unique gait, leading to advantages in functional and patient reported outcomes over the control and baseline conditions.

Recruitment & Eligibility

Status: COMPLETED

Sex: All

Target Recruitment: 10

Inclusion Criteria

A unilateral amputation of the lower limb
Aged between 18 to 75 years, inclusive
K3 or K4 level ambulators who can perform all locomotor tasks of interest (based on assessment of the physiatrist and/or prosthetist)
If a prosthesis is used, the participant must use a prosthetic knee and foot in their clinically prescribed prosthesis.

Exclusion Criteria

Individuals with history of neurological injury, gait pathology, or cardiovascular condition that would limit ability to ambulate for multiple hours
Individuals who are currently pregnant (based on patient self-report) due to slight risk of falling during experiments

Study & Design

Study Type: INTERVENTIONAL

Study Design: SINGLE_GROUP

Arm && Interventions

Group	Intervention	Description
Smart Robotic Knee/Ankle Prothesis	Robotic Knee/Ankle Prosthesis	This study will be conducted on a sample population of individuals with transfemoral amputation (single arm). Each participant will test with each condition of the study (repeated measures).

Primary Outcome Measures

Name	Time	Method
Overground Self-selected Walking Speed	1 day	This measures the individuals preferred overground walking speed which indicates their physical capability with a device.

Secondary Outcome Measures

Name	Time	Method
Overground Walking Speed Mean Absolute Error (MAE)	1 day	This outcome is the error with which the machine learning model embedded into our advanced prosthesis controller's microprocessor predicts the user's walking speed overground. Specifically, mean absolute error (MAE) is computed between the predicted walking speed and the ground truth walking speed, or the speed that the user is actually walking at. Ground truth measurements are measured by a motion-capture system and taken to be center of mass speed. Walking speed predictions are made every 50 ms and compared to the nearest center-of-mass speed. For this measure, lower walking speed MAEs are indicative of greater accuracy in defining the user's true walking speed and thus lower numbers are indicative of an improved outcome.
Treadmill Walking Speed Mean Absolute Error (MAE)	1 day	This outcome is the error with which the machine learning model embedded into our advanced prosthesis controller's microprocessor predicts the user's walking speed on the treadmill. Specifically, mean absolute error (MAE) is computed between the predicted walking speed and the ground truth walking speed, or the speed that the user is actually walking at. Ground truth measurements are measured by the true treadmill speed (for treadmill trials). Walking speed predictions are made every 50 ms and compared to the nearest center-of-mass speed. For this measure, lower walking speed MAEs are indicative of greater accuracy in defining the user's true walking speed and thus lower numbers are indicative of an improved outcome.