Study at a glance
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: This review reports on studies using Theia3D versions ranging from v2020.3.0.962 to v2023.1.0.3161, though six of the sixteen included studies did not report which version they used.
Study Overview
Article Type and Scope
A systematic review following PRISMA guidelines, synthesizing the published literature on the accuracy, validity, and reliability of Theia3D markerless motion capture across functional tasks and participant populations, rather than reporting new data collection.
Included Studies and Populations
Sixteen cross-sectional studies met inclusion criteria out of 1,191 records initially identified, involving a total of 303 participants: 180 adults (85 female, 95 male), including 46 with a health-related movement condition such as knee osteoarthritis, and 123 children, including 66 diagnosed with a condition such as cerebral palsy, cavovarus foot, Charcot-Marie-Tooth disease, spina bifida, or a spinal arteriovenous malformation. Studies were required to compare Theia3D against a validated three-dimensional reference method, such as marker-based motion capture or inertial measurement units, and to report at least one biomechanical outcome measure in a sample of more than four participants.
Methodology and Quality Assessment
Two reviewers independently screened and extracted data from included studies, assessing methodological quality using a modified version of the COSMIN (COnsensus-based Standards for the selection of health Measurement INstruments) risk-of-bias checklist, alongside two additional criteria evaluating content validity and the transparency of Theia3D's underlying AI methodology. The overall quality of evidence for each pooled outcome was then rated using a modified GRADE (Grading of Recommendations Assessment, Development, and Evaluation) framework, which downgrades evidence quality based on risk of bias, imprecision (driven by sample size), and content validity concerns.
Key Findings
- Lower-limb kinematics during gait showed reasonable overall agreement with marker-based systems in healthy adults, with hip rotation standing out as the least consistent measure. The smallest differences occurred for hip abduction/adduction (root-mean-square differences as low as 2.6°, indicating excellent agreement), while hip flexion/extension (10° to 11°) and internal/external rotation (6.9° to 13.6°) showed the largest differences, corresponding to poor agreement by the review's accuracy criteria.
- Spatiotemporal gait parameters showed excellent validity and inter-session reliability, rated as high-quality evidence overall. Gait speed, step length, and stride length showed intraclass correlation coefficients above 0.9 and standard errors of measurement at or below 0.07 m/s for speed and 0.06 m for step and stride length, while swing time, stance time, and double-limb support time showed more variable, moderate-to-poor agreement.
- Evidence in children with movement-related conditions was rated very low quality and showed poorer agreement than in healthy adults. In children with cerebral palsy specifically, trunk, pelvis, and hip flexion/extension angles showed differences ranging from about 5° to 12°, and hip and knee rotation differences reached into the 9° to 13° range, considerably larger than the differences observed for the same joints and planes in healthy adult gait.
- Running and jumping kinematics were assessed in fewer studies and were generally rated lower quality evidence than gait. During running, hip rotation again showed the largest differences (9.8° to 12.8°), while during jumping, hip flexion/extension showed root-mean-square differences as high as 20.6°, the largest single discrepancy reported anywhere in the review, with the overall quality of evidence for jumping rated low.
- Inter-session reliability for lower-limb gait kinematics was considered low-quality evidence but showed reasonable consistency in the frontal and sagittal planes, with average inter-session variability ranging from about 1.5° to 2.8° across the studies examined, and somewhat greater variability consistently observed in the transverse plane than in the frontal or sagittal planes.
What This Means for Movement Science Researchers and Lab Methodologists
For researchers deciding where Theia3D's current evidence base is strong enough to support confident use, this review offers a plane- and parameter-specific answer rather than a single verdict: spatiotemporal gait parameters and frontal- and sagittal-plane lower-limb kinematics during gait are supported by comparatively strong evidence, while hip flexion/extension, transverse-plane rotations, and running and jumping tasks remain areas where the published evidence is thinner and less consistent. The review's finding that evidence quality was consistently lower in children with movement-related conditions than in healthy adults is a particularly important signal for teams working in pediatric or other populations with movement conditions, underscoring the authors' call for more, higher-quality studies specifically in those groups before markerless kinematic data from these populations can be interpreted with the same confidence as healthy-adult data.
The Role of Theia3D in Movement Analysis
Theia3D was the system under review throughout this paper, with the authors pooling accuracy, validity, and reliability findings from sixteen independently conducted studies that had each compared Theia3D against a marker-based or inertial reference system across a range of tasks and populations. Because this is a review of previously published findings rather than new data collection, Theia3D's own role here was as the common subject across all sixteen included studies, allowing the authors to identify where its published evidence is most consistent and where meaningful gaps remain.
Study Limitations and Future Directions
The authors note that six of the sixteen included studies did not report which version of Theia3D they used, which limits the ability to track whether software updates have meaningfully changed accuracy or reliability over time, and they recommend that future studies document software versioning as standard practice. Content validity was inconsistently reported across studies, with most failing to document camera shutter speed or lighting conditions, both considered important parameters that can influence markerless tracking performance. The commercial, proprietary nature of Theia3D's underlying AI model means there is limited publicly available information about its training data and landmark identification methods, which the authors argue limits researchers' ability to fully assess the model's suitability for specific populations or movement types not well represented in its training data. The included studies used a relatively small number of participants overall, with about a third of studies involving fewer than 21 participants, and differences in the biomechanical models and comparison methods used across studies made some findings difficult to directly compare or pool. The authors identify standardizing biomechanical model definitions and comparison methodologies, along with conducting more rigorous studies specifically in populations with movement-related conditions, as clear priorities for strengthening the evidence base going forward.
Citation: Varcin, F., & Boocock, M. G. (2026). The accuracy, validity and reliability of Theia3D markerless motion capture for studying the biomechanics of human movement: A systematic review. Artificial Intelligence In Medicine, 173, 103332. https://doi.org/10.1016/j.artmed.2025.103332
Link: Read the full study
What Does the Full Body of Evidence Say About Theia3D?
This systematic review pools sixteen independently conducted studies to show where Theia3D's published accuracy, validity, and reliability evidence is strongest, spatiotemporal gait parameters and frontal/sagittal-plane kinematics, and where more research is still needed, hip rotation, transverse-plane motion, and populations with movement-related conditions. If your work involves understanding the full evidence base behind markerless motion capture for your own research or assessment context, talk to our team about what this review could mean for your application.