Multiscale 3D shape representation and segmentation with applications to hippocampal/caudate extraction from brain...

Area of Science:

• Medical image analysis
• Computer vision
• Computational anatomy

Background:

• Medical image segmentation is crucial but challenging due to image quality and complex structures.
• Existing shape-based methods struggle with limited shape variations and loss of local shape information.
• The need for robust segmentation methods that can handle arbitrary topologies and small shape variances is critical.

Purpose of the Study:

• To develop a fully automatic medical image segmentation method using a novel multiscale shape representation.
• To overcome limitations of previous shape-based approaches by enriching shape variations and preserving local details.
• To improve segmentation accuracy for challenging medical datasets with low contrast and sharp features.

Main Methods:

• A multiscale shape representation is created using wavelet transforms to capture shape variances at various scales.
• Statistical learning is applied to represent shape variations across different scales, enriching eigen-shapemodes.
• A multi-atlas initialization procedure incorporates both shape and grayscale image information.
• Segmentation is performed by combining multi-atlas initialization with the multiscale shape knowledge.

Main Results:

• The multiscale shape representation effectively captures small-scale shape changes and enriches eigen-shapemodes.
• The proposed method demonstrates statistically significant improvements in segmentation tasks.
• Successful segmentation of challenging medical datasets with low contrast and sharp corner structures was achieved.

Conclusions:

• The developed multiscale shape representation significantly enhances the ability to represent complex shapes.
• The proposed segmentation method offers a statistically significant improvement over existing techniques for challenging medical images.
• This approach provides a more robust and accurate solution for medical image segmentation, particularly for structures with intricate details.