Biplanar Videoradiography Dataset for Model-based Pose Estimation Development and New User Training.

Area of Science:

• Biomechanics
• Medical Imaging
• Orthopedics

Background:

• Accurate measurement of in vivo bone motion is crucial for understanding foot and ankle pathologies.
• Biplanar videoradiography is a suitable technique for in vivo bone motion analysis, but pose estimation remains challenging.
• Current model-based pose estimation methods lack accuracy due to user dependency and lab-specific algorithms.

Purpose of the Study:

• To present a novel dataset of in vivo calcaneus, talus, and tibia poses during dynamic activities (running, hopping).
• To introduce and validate a training method using marker-based visual feedback to improve model-based pose estimation accuracy.
• To provide a dataset for validating other model-based pose estimation software.

Main Methods:

• Collected a rare in vivo dataset of calcaneus, talus, and tibia poses using marker-based methods during running and hopping.
• Developed a training protocol to enhance user accuracy in model-based pose estimation software.
• Utilized marker-based visual feedback to guide users in refining their initial pose estimations.

Main Results:

• New operators, after training, achieved bone pose estimations within 2° of rotation and 1 mm of translation compared to marker-based ground truth.
• The achieved accuracy is comparable to expert users and significantly reduces inter-operator variability.
• The presented dataset enables validation of alternative model-based pose estimation techniques.

Conclusions:

• The developed training method substantially improves the accuracy and consistency of in vivo bone pose estimation using biplanar videoradiography.
• The shared dataset serves as a valuable resource for validating and advancing motion analysis techniques in biomechanics and orthopedics.
• This work facilitates faster and more accurate bone pose measurements, aiding in the diagnosis and understanding of musculoskeletal disorders.