Personalizing powered knee prostheses using reinforcement learning

Towards Efficient Personalization of Computerized Lower Limb Prostheses Via Reinforcement Learning in a Clinical Setup - Patient Study

This trial will test whether a computer system called RISE can help clinicians tune powered knee prostheses faster and make them work better for people with above-knee (transfemoral) amputations.

Quick facts

• Phase: PHASE1
• Study type: Interventional
• Enrollment: 24 (estimated)
• Ages: 18 Years to 75 Years
• Sex: All
• Sponsor: North Carolina State University (other)
• Locations: 1 site (Raleigh, North Carolina)
• Trial ID: NCT07204925 on ClinicalTrials.gov

What this trial studies

Researchers will compare a reinforcement-learning based Recommendation Interfacing System (RISE) against standard manual tuning to personalize an experimental powered prosthetic leg for unilateral transfemoral amputees. Participants will be fitted with a prototype robotic knee and randomized to personalization guided by RISE (performed by tuning experts or prosthetists) or to manual personalization by tuning experts. The primary measures are clinician tuning speed and short-term patient functional performance during walking tasks. This Phase 1 interventional trial focuses on feasibility and early functional outcomes in a clinical setting.

Who should consider this trial

Good fit: Ideal candidates are unilateral transfemoral amputees age 18–75 who are at least one year post-amputation, K-level 3 or higher, using a stable socket and leg for more than three months, can walk at least four minutes without assistive devices, and have no major recent residual-limb skin problems.

Not a fit: Patients with very short residual thighs, extreme body size (height <1.50 m or weight >116 kg), significant cognitive/visual/hearing impairments, recent major skin issues, or medical comorbidities that affect safety are unlikely to benefit or be eligible.

Why it matters

Potential benefit: If successful, the system could shorten fitting time, allow less-specialized clinicians to perform effective tuning, and improve walking performance for amputees.

How similar studies have performed: Laboratory and simulation work has shown promise for algorithmic and adaptive controllers, but clinical use of reinforcement-learning guided personalization for powered prostheses is still novel with limited clinical trial evidence.

Eligibility criteria

Inclusion Criteria:

* Unilateral transfemoral amputees between 18-75 with K level three or higher
* More than one year after amputation
* Using current prosthetic socket and leg for more than three months
* No major skin issues on the residual limb for more than six months
* Can walk for more than 4 minutes continuously without any other assistive devices

Exclusion Criteria:

* Have very short residual thighs (the length of the residual limb is \<15% of the length of the unimpaired limb)
* Are \<1.50m in height or \>116Kg in weight (who would not fit our prosthesis or the PowerKnee)
* Have cognitive, visual, audio impairments that would affect their ability to give informed consent or to follow simple Instructions during the experiments
* Have any significant co-morbidity that interferes with the study (e.g. stroke, pacemaker placement, pain, etc.).

Where this trial is running

Raleigh, North Carolina

• North Carolina State University — Raleigh, North Carolina, United States (RECRUITING)

Study contacts

• Principal investigator: He (Helen) Huang, PhD — NC State University
• Study coordinator: Ming Liu, PhD

Conditions: Transfemoral Amputation, Powered prosthesis legs, control parameter personalization, reinforcement learning