Training models of anatomic shape variability.

Area of Science:

• Medical image analysis
• Computational anatomy
• Statistical shape modeling

Background:

• Accurate statistical segmentation and registration in medical imaging rely on high-quality shape fitting.
• Current methods for shape fitting are often ad hoc and overlook general principles.
• The quality of statistical analysis is directly tied to the initial shape fitting process.

Purpose of the Study:

• To present general principles for training shape models of anatomical structures.
• To introduce a novel method for jointly estimating geometric models and shape distributions.
• To improve the accuracy and robustness of statistical segmentation and registration methods.

Main Methods:

• A novel iterative approach to jointly estimate geometric models and shape distributions.
• Relaxing geometric constraints in favor of converging shape probabilities.
• Crafting geometric constraints for legal, non-self-intersecting shapes and establishing model-to-model correspondences.

Main Results:

• Demonstrated application to synthetic and real patient CT image datasets.
• Successful application to diverse anatomical structures including pelvis, head and neck, kidneys, and brain.
• Validation of the method's effectiveness in training shape models for medical imaging.

Conclusions:

• The proposed method provides a principled approach to shape model training.
• This shape training is foundational for image-guided radiation therapy (IGRT) and adaptive radiation therapy (ART).
• The method enhances the reliability of statistical segmentation and registration in medical image analysis.