Study at a glance

In 20 healthy adults performing six shoulder tasks through the full range of arm motion, Theia3D agreed closely with marker-based motion capture for abduction angles in the coronal plane, but showed larger, practically meaningful differences for flexion and rotation angles, especially at the extremes of each movement.

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 v2024.1.24, as stated in the paper.

Study Overview

Design

A within-subject comparison of shoulder kinematics measured simultaneously by Theia3D markerless motion capture and a marker-based motion capture system, across six shoulder tasks spanning the full range of arm motion in three planes.

Participants

Twenty healthy young adults (10 men, 10 women; mean age 27.1) with no known elbow or shoulder conditions, neuromuscular disorders, or other conditions affecting upper-limb movement. Both shoulders were analyzed for each participant, giving 40 shoulders in total.

Movement Tasks

Each participant performed bilateral scaption, abduction, flexion, internal rotation, external rotation, and internal and external rotation at 90 degrees of abduction, three repetitions of each, seated, with full verbal and visual task instructions.

Data Collection

Synchronized markerless (Qualisys Miqus cameras, 100 Hz) and marker-based (Qualisys Arqus cameras, 200 Hz) motion capture recorded each task simultaneously. Participants wore anatomical reflective markers on the upper extremities and torso following International Society of Biomechanics guidelines; markerless data were processed in Theia3D and both datasets were further processed in Matlab.

Analysis

Humerothoracic joint angles from the markerless system were compared against the marker-based reference using root-mean-square difference (RMSD) and coefficient of multiple correlation (CMC) for each angle and task. Paired t-tests assessed differences in range of motion between systems, and results were weighed against minimal practically important differences established in the shoulder-kinematics literature.

Key Findings

  • Abduction angles in the coronal plane showed the closest agreement between systems. RMSDs for abduction ranged from 5.3° to 6.0° with excellent correlations, the smallest differences and highest correlations of any plane measured.
  • External rotation and other transverse-plane angles showed the largest disagreement. RMSDs for external rotation ranged from 15.3° to 22.3° depending on the task, well above the study's minimal practically important difference threshold, with correlations ranging from weak to good depending on the specific task.
  • Range-of-motion measurements differed significantly between systems for nearly every task in nearly every plane. Significant differences appeared in the coronal plane for the flexion and external rotation tasks, in the sagittal plane for all tasks except internal rotation and external rotation at 90-degree abduction, and in the transverse plane for every task analyzed.
  • Despite these statistically significant differences, most mean range-of-motion gaps between systems stayed within the practically meaningful threshold. Coronal- and sagittal-plane range-of-motion differences were generally within the minimal practically important difference, even where they reached statistical significance; transverse-plane differences, in contrast, exceeded that threshold for several tasks, reaching as much as 43.2° for rotation at 90-degree abduction.
  • Disagreement between systems was consistently largest near the extremes of each movement's range. Both angle waveforms and range-of-motion comparisons showed the biggest gaps between markerless and marker-based data toward the inflection points of each task, where the arm approached its maximum reach.

What This Means for Movement Science Researchers and Teams

For teams considering markerless capture for shoulder assessment, this study offers a plane-specific picture rather than a single verdict: Theia3D currently performs well enough in the coronal plane, especially for abduction, to be a credible alternative to marker-based capture, but sagittal- and transverse-plane shoulder kinematics still show gaps large enough to warrant caution before real-world or training application. The authors note that Theia3D does not model the scapula, which may limit how precisely it can characterize humerothoracic motion, and that the version-to-version improvement they observed for lower-limb kinematics did not carry over as strongly to shoulder measurements, a reminder that a markerless system's accuracy for a given joint depends heavily on how well-represented that joint and its movement range are in the system's training data.

The Role of Theia3D in Movement Analysis

Theia3D was directly evaluated throughout this study, with its humerothoracic joint angles compared against a simultaneously recorded marker-based reference across the full range of shoulder motion in all three planes. Because markerless capture requires no marker placement or preparation time, this kind of full-range shoulder assessment is substantially faster to set up than an equivalent marker-based protocol, a practical advantage the authors note could support both training and recovery progress monitoring once accuracy across all planes is further established.

Study Limitations and Future Directions

The authors note several limitations. Although marker placement was consistently performed by the same person with palpation skill, some variability in marker placement cannot be excluded, and marker-based measurements themselves do not represent a true, error-free reference of shoulder kinematics due to skin movement artefacts. Because the markerless system is sensitive to video quality, results could differ under less controlled lighting, contrast, or clothing conditions than those used in this study. The study included only healthy young adults without known shoulder or elbow disorders, so findings may not generalize to individuals with shoulder-specific conditions, and results could differ in people with a higher body mass index or other factors affecting upper-limb kinematics. The authors identify further Theia3D development, particularly additional training data for sagittal- and transverse-plane shoulder motion, as the clear next step toward the full three-plane accuracy that real-world and performance applications would require.

Citation: Croci, E., Künzel, C., Gehring, D., Müller, A. M., & Mündermann, A. (2025). Comparison of shoulder kinematics between Theia 3D markerless motion capture and marker-based motion capture during full arm ranges of motion. Journal of Biomechanics, 191, 112905. https://doi.org/10.1016/j.jbiomech.2025.112905

Link: Read the full study

How Far Does Markerless Shoulder Accuracy Really Extend?

This study shows that Theia3D's accuracy for shoulder kinematics isn't uniform across planes of motion, strong for abduction, less consistent for flexion and rotation, especially near the limits of a movement's range. If your work involves shoulder assessment and you want to understand exactly where markerless capture is validated for upper-limb movement today, talk to our team about what this means for your specific application.

Recent Posts