Particle system based adaptive sampling on spherical parameter space to improve the MDL method for construction of...

Area of Science:

• Medical imaging
• Computer vision
• Computational geometry

Background:

• Statistical shape models (SSMs) are crucial for analyzing anatomical variations.
• Minimum description length (MDL) based group-wise registration is a standard technique for creating SSMs.
• Existing MDL methods struggle with non-uniform point distribution on 3D shapes.

Purpose of the Study:

• To develop an adaptive sampling method for improved point correspondence in 3D shape registration.
• To address the non-uniformity issue in point distribution inherent in traditional MDL-based registration for SSM construction.

Main Methods:

• Proposed a particle-system approach on a unit sphere to achieve adaptive sampling positions.
• Minimized particle system energy related to mesh distortions for optimal sampling.
• Constructed a spherical mesh from steady-state particles and relaxed it for precise vertex placement.

Main Results:

• The particle-system method successfully resolved the non-uniform sampling problem of the original MDL approach.
• Evaluated on lung, liver, kidney, and spleen datasets, demonstrating superior performance.
• Achieved improved results in generalization and specificity tests compared to the baseline MDL method.

Conclusions:

• The proposed particle-system based method offers a significant advancement in 3D shape registration for SSM construction.
• This adaptive sampling technique leads to more robust and accurate statistical shape models.
• The method shows promise for various anatomical structures in medical image analysis.