Joint Segmentation and Shape Regularization With a Generalized Forward-Backward Algorithm.

Area of Science:

• Medical image analysis
• Computational geometry
• Computer vision

Background:

• Accurate segmentation and regularization of shape sequences are crucial for analyzing dynamic 3D data.
• Existing methods often struggle with complex shape variations and computational efficiency.

Purpose of the Study:

• To develop a robust and efficient method for simultaneous segmentation and regularization of shape series from image sequences.
• To apply this method to 3D data and extend it for 3D+t imaging modalities.

Main Methods:

• A novel model incorporating total variation prior on the shape manifold.
• Utilizing a modified Kendall shape space and Sobolev gradients for explicit computations.
• An efficient splitting scheme based on a generalized forward-backward approach with proximal mappings and gradient descent.

Main Results:

• Demonstrated effectiveness on various 3D data application examples.
• Successful extension to shape fields for 3D+t imaging modalities.
• The method provides accurate segmentation and regularization of shape sequences.

Conclusions:

• The proposed method offers an efficient and computationally accessible solution for shape analysis in 3D and 3D+t imaging.
• This approach advances the field of medical image analysis by enabling more precise tracking and regularization of anatomical structures over time.