Deformable segmentation via sparse shape representation.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Biomedical Engineering

Background:

• Medical image segmentation relies on appearance and shape, but weak or misleading appearance cues from disease or artifacts can cause errors.
• Existing methods struggle with segmentation when appearance information is compromised, necessitating more robust approaches.

Purpose of the Study:

• To propose a novel deformable model for robust medical image segmentation, particularly when appearance cues are weak or misleading.
• To improve segmentation accuracy by focusing on advanced shape modeling techniques.

Main Methods:

• Developed a novel deformable model emphasizing shape composition for on-the-fly incorporation of shape priors, robust to false appearance information.
• Implemented a hierarchical shape prior modeling approach using affinity propagation to partition surfaces and build local shape models independently.
• Applied the deformable model to liver segmentation in PET-CT images and rodent brain segmentation in MR images.

Main Results:

• The proposed shape composition method demonstrates robustness against false appearance information and adaptability to minor shape variations.
• Hierarchical modeling using affinity propagation enables compact and efficient shape prior representation, leading to improved segmentation.
• The deformable model achieved superior performance in segmenting both liver in PET-CT and rodent brains in MR images compared to state-of-the-art methods.

Conclusions:

• The novel deformable model offers a robust solution for medical image segmentation challenges posed by weak or misleading appearance cues.
• Effective shape modeling, through on-the-fly composition and hierarchical representation, is crucial for accurate segmentation in complex scenarios.
• The method's successful application in diverse imaging modalities (PET-CT, MRI) highlights its versatility and effectiveness.