Entropy-based Correspondence Improvement of Interpolated Skeletal Models

Area of Science:

• Medical imaging analysis
• Computational anatomy
• Statistical shape modeling

Background:

• Statistical analysis of shape representations requires accurate correspondence across populations.
• Point distribution models (PDMs) are commonly used for object boundary representation.
• Skeletal representations (s-reps) offer richer information (width, direction, position) than boundary-based PDMs, potentially improving correspondence.

Purpose of the Study:

• To develop and evaluate methods for improving correspondence in skeletal models.
• To introduce continuous interpolation for discretely-sampled skeletal models.
• To enhance statistical shape analysis through improved s-rep correspondence.

Main Methods:

• Developed a continuous interpolation method for skeletal models based on medial structures.
• Implemented a correspondence improvement technique by shifting skeletal samples using interpolation.
• Evaluated correspondence using regularity and shape-feature population entropies on synthetic and real data.

Main Results:

• The proposed interpolation method successfully refines skeletal models.
• The correspondence improvement technique enhanced the alignment of s-rep models.
• Optimized s-rep models and combined boundary-and-skeletal PDMs showed superior correspondence compared to boundary-only PDMs.
• Demonstrated improved correspondence for s-rep models fitted to segmented lateral ventricles.

Conclusions:

• Continuous interpolation of skeletal models significantly improves correspondence.
• The developed methods enhance the accuracy of statistical shape analysis.
• Skeletal representations offer advantages over boundary-based methods for shape analysis, particularly in medical applications.