Multi-atlas Based Segmentation Editing with Interaction-Guided Constraints.

Area of Science:

• Medical image analysis
• Computer vision
• Computational anatomy

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• Existing multi-atlas methods often rely heavily on appearance features, limiting their effectiveness in editing scenarios.
• Manual segmentation refinement can be time-consuming and requires specialized expertise.

Purpose of the Study:

• To develop a novel multi-atlas based segmentation method for image editing tasks.
• To incorporate user interaction-guided constraints to improve segmentation accuracy and efficiency.
• To provide a computationally inexpensive and broadly applicable solution for segmentation editing.

Main Methods:

• A multi-atlas segmentation approach incorporating interaction-guided constraints.
• User interactions on erroneous segmentation regions are used to guide the selection of training labels.
• Local search for matching training label patches based on interaction combinations and previous segmentation.
• Label fusion of selected patches with weights determined by distance to interactions.

Main Results:

• The proposed method successfully edited segmentations across challenging datasets (prostate CT, brainstem CT, hippocampus MR).
• It demonstrated superior performance compared to existing image editing methods on all tested datasets.
• The method effectively handles various shape changes with limited training data and user interactions.

Conclusions:

• The novel interaction-guided multi-atlas method offers an effective solution for segmentation editing.
• It provides a computationally efficient alternative to traditional methods, requiring no image information or extensive learning.
• The approach shows significant potential for improving clinical workflows in medical image analysis.