Human Movement & Mobility Research
Not specified

Markerless 3D Gait Analysis in Individuals with Lower-Limb Amputation in Routine Care: Spatiotemporal, Kinematic, and Kinetic Patterns According to the Level of Amputation

Journal of Prosthetics and Orthotics
•
2026
•
Original research
Theia3D version:
v2025.1.2
More Articles

Study at a glance

In 39 people with lower-limb amputation assessed during routine prosthetic care, Theia3D captured gait patterns consistent with decades of marker-based amputee gait literature, while also surfacing an unexpected, biomechanically implausible pattern in prosthetic ankle motion that the authors flag as a specific validity concern.

Summary

Disclaimer: The following summary details independent academic research. Theia3D is intended for research, sports performance and educational use only. It is not cleared as a medical device and is not intended to be used for the diagnosis, treatment, mitigation or prevention of any disease, injury or medical condition.

Editor's note: Theia3D version v2025.1.2, as stated in the paper. The paper does not name a specific camera brand/model for the markerless system; it reports only that an 11-camera system was used.

Study Overview

Design

A single-center, prospective study comparing markerless gait data from people with transtibial and transfemoral amputation, collected during routine outcome-tracking appointments, against an existing marker-based reference dataset from typically developed adults.

Participants

Nineteen participants with unilateral transtibial amputation (median age 58, 84% male) and 20 with unilateral transfemoral amputation (median age 52, 95% male) were included, all using a functional prosthesis and able to walk on a level surface. The most common causes of amputation in both groups were trauma and tumor, and median time since amputation was roughly 10 to 12 years. One transtibial and four transfemoral participants used forearm crutches during testing. A reference dataset of 21 typically developed adults (median age 27.7), drawn from an existing institutional database and assessed separately using marker-based motion capture, served as the comparison group.

Data Collection

Participants with amputation walked at a self-selected pace along a 7-meter walkway for at least 10 trials, wearing shoes, while an 11-camera system integrated into a standard motion-capture setup recorded at 60 Hz and was processed through Theia3D. Ground reaction forces were recorded synchronously via three force plates; gait events were automatically detected from the force data and visually corrected when needed. The reference group's marker-based data had been collected separately using a 12-camera system at 120 Hz with a standard lower-body marker set plus additional trunk markers.

Analysis

Kruskal-Wallis tests compared spatiotemporal, kinematic, and kinetic variables across the three groups, with Dunn's test and Bonferroni correction for pairwise follow-up comparisons where differences were found. Lower-limb analyses focused on the prosthetic limb specifically, since this represented the greatest challenge for the markerless software.

Key Findings

  • Spatiotemporal parameters showed the clearest and most consistent markerless gait signature in this population. Both the transtibial and transfemoral groups walked slower, with longer stride time, shorter stride length, wider steps, and more time in double support than the typically developed group, and transfemoral participants showed more pronounced changes (lower cadence, longer step time) than transtibial participants, matching patterns long established in the marker-based literature.
  • Sagittal-plane kinematics and kinetics at the hip and ankle also matched expected prosthetic gait patterns. Both amputation groups showed reduced ankle plantarflexion range of motion and push-off moment, reduced knee flexion during loading response, and a trend toward reduced hip extension paired with a significantly reduced hip extension moment, all consistent with established features of prosthetic gait.
  • The data suggested a biomechanically implausible pattern of ankle motion during the swing phase. Markerless tracking showed the prosthetic ankle appearing to move in a way that resembles natural dorsiflexion during swing, a movement a passive prosthetic foot cannot physically produce on its own. The authors suggest the algorithm, trained primarily on people without amputation, may be applying its learned expectation that a "foot" dorsiflexes during swing to a prosthetic foot it doesn't recognize as different, though they're clear this explanation is speculative rather than confirmed, since the underlying model's decisions aren't visible to the researchers.
  • Pelvic kinematics were less consistent with the literature than other measures. Pelvic tilt range of motion was higher in both amputation groups than in the comparison group, but no difference was found between transtibial and transfemoral participants despite that being reported elsewhere, which the authors link to markerless capture's known tendency to underestimate pelvic tilt, a limitation that can also affect how hip flexion is calculated downstream.
  • Frontal-plane hip and trunk findings aligned more closely with expectations, particularly in the transfemoral group. Both groups showed reduced hip abduction and adduction range of motion and reduced hip adduction moment, while increased trunk tilt and obliquity appeared primarily in the transfemoral group, consistent with greater trunk involvement at a higher level of amputation.

What This Means for Prosthetics Researchers and Care Teams

For anyone considering markerless capture as a way to add objective gait data to a prosthetic outcomes program, this study delivers a genuinely mixed message. From the participant's side, data collection was fast and well tolerated, taking roughly 10 to 15 minutes compared with about 45 minutes for the broader outcomes assessment it was added to. But two things currently stand in the way of routine use: post-processing took up to two hours per person, workable in a research context but not yet practical without dedicated staff and workflow changes, and the implausible ankle-motion finding means markerless outcomes for this population haven't yet been specifically validated against a simultaneous marker-based reference. The authors are explicit that, until that validation exists, these markerless outcomes shouldn't be used on their own to inform decisions about an individual's prosthetic management or progress, even though the broader spatiotemporal and sagittal-plane findings here are reassuring.

The Role of Theia3D in Movement Analysis

Theia3D processed video from an 11-camera system to generate the spatiotemporal, kinematic, and kinetic gait data at the center of this study, without requiring markers or the setup time that can be especially disruptive in a routine care appointment. At the same time, this study is a useful case study in where markerless capture's training data can matter: because Theia3D's underlying model has been trained mostly on people without limb loss, this is likely at least part of why it struggled specifically with motion at the prosthetic ankle, a structure that looks and moves differently from the anatomical joint the model was built to recognize, while performing well on spatiotemporal and most other kinematic measures that are less dependent on recognizing the specific anatomy of a given joint.

Study Limitations and Future Directions

The comparison group was notably younger than both amputation groups, so some age-related effects on spatiotemporal parameters can't be ruled out, and the transfemoral group was 95% male, meaning sex-related effects couldn't be separated from group effects. Because participants were recruited during ordinary care appointments rather than for a controlled study, the sample was heterogeneous in prosthetic components, time since amputation, and walking-aid use, which reflects real-world variation well but may have reduced the study's ability to detect group-level effects, and the small number of walking-aid users couldn't be evaluated separately. The comparison group itself was assessed with marker-based rather than markerless capture, so this study wasn't designed as a direct validation against a marker-based reference in the same participants, true validation would require simultaneous marker-based and markerless measurement in people with amputation specifically, which the authors note is logistically difficult to arrange during routine appointments. Clothing also wasn't standardized, and how clothing and prosthetic cosmetic covers specifically affect markerless tracking in this population remains unclear. The authors frame simultaneous marker-based and markerless validation in this population as the clear next step, particularly to confirm or rule out the implausible ankle-motion finding.

Citation: Zeitzschel, A., Salehi Tabar, Z., Heitzmann, D. W. W., Bisele, M., Alimusaj, M., & Wolf, S. I. (2026). Markerless 3D gait analysis in individuals with lower-limb amputation in routine care: Spatiotemporal, kinematic, and kinetic patterns according to the level of amputation. Journal of Prosthetics and Orthotics. (Volume and issue not yet assigned at the time of this summary.)

Link: Read the full study

What Would It Take to Trust Markerless Gait Data for Prosthetic Users?

This study shows markerless capture getting most of amputee gait right while catching itself on a genuinely surprising edge case at the prosthetic ankle, exactly the kind of finding that matters for knowing where a system can and can't be trusted yet. If your research involves gait in people with limb loss and you want to understand where the current validation evidence stands, talk to our team about what's been tested in this population and what hasn't.

Recent Posts