Geodesic patch-based segmentation.

Area of Science:

• Medical image analysis
• Computer-aided diagnosis
• Computational anatomy

Background:

• Label propagation is effective for automatic segmentation but sensitive to image registration errors.
• Patch-based methods reduce registration dependence but are limited by search window size, affecting patch selection for label fusion.

Purpose of the Study:

• To develop a novel patch-based label propagation approach that overcomes limitations of existing methods.
• To improve the accuracy and robustness of medical image segmentation by incorporating patient-specific spatial context.

Main Methods:

• A novel patch-based label propagation approach utilizing relative geodesic distances.
• Definition of patient-specific coordinate systems to provide spatial context.
• Evaluation on multi-organ segmentation tasks using cardiac MR and abdominal CT images.

Main Results:

• The proposed method demonstrates competitive results in multi-organ segmentation.
• The approach effectively addresses limitations related to registration errors and search window size in patch-based methods.

Conclusions:

• The novel patch-based label propagation method offers a robust solution for medical image segmentation.
• Incorporating relative geodesic distances for patient-specific coordinate systems enhances segmentation accuracy and reliability.