Automatic generation of statistical pose and shape models for articulated joints

Area of Science:

• Medical imaging
• Biomechanical analysis
• Computational anatomy

Background:

• Statistical motion modeling of joints aids in pathology detection but faces challenges with complex joints like the wrist.
• Accurate registration and segmentation of multi-subject 3-D volumes at various articulated positions are crucial for robust modeling.

Purpose of the Study:

• To present a novel framework for simultaneous registration and segmentation of 3-D wrist volumes from multiple subjects.
• To develop an automated method for generating statistical pose and shape models of carpal bones.
• To create a measurement tool for diagnosing scaphoid-lunate dissociation using the developed models.

Main Methods:

• A novel framework for simultaneous registration and segmentation of multiple 3-D (CT/MR) volumes from different subjects at various articulated positions.
• Utilized an initial pose model from a template subject, refined iteratively with the Grow-Cut algorithm for bone segmentation and pose parameter estimation.
• Updated the pose model with each newly registered and segmented subject to enhance successive registration accuracy.

Main Results:

• The framework demonstrated robust and accurate performance on CT images of 25 subjects, with an average mean target registration error of 0.34 ±0.27 mm.
• Automatic segmentation results showed high consistency with semi-automatically obtained ground truth.
• The generated statistical pose and shape models enabled a diagnostic tool for scaphoid-lunate dissociation with 90% sensitivity and specificity.

Conclusions:

• The proposed framework successfully enables simultaneous registration and segmentation of multi-subject 3-D wrist volumes.
• The automated generation of statistical pose and shape models is feasible and accurate.
• The developed models are valuable for creating effective diagnostic tools for wrist pathologies like scaphoid-lunate dissociation.