INTEGRATING SEMI-SUPERVISED LABEL PROPAGATION AND RANDOM FORESTS FOR MULTI-ATLAS BASED HIPPOCAMPUS SEGMENTATION

Area of Science:

• Medical image analysis
• Computer vision
• Machine learning

Background:

• Accurate segmentation of anatomical structures like the hippocampus in medical images is crucial for diagnosis and treatment planning.
• Existing multi-atlas based segmentation methods face challenges in achieving high accuracy and robustness.

Purpose of the Study:

• To propose a novel and improved multi-atlas based image segmentation method.
• To enhance the accuracy and reliability of hippocampus segmentation in MR images.

Main Methods:

• Integration of a semi-supervised label propagation method with a supervised random forests method within a label fusion framework.
• Utilizing random forests to train a regression model on atlas image patches for voxel-wise segmentation guidance.
• Employing label propagation that considers both local and global image appearance for segmentation.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art multi-atlas segmentation techniques.
• Experimental results validated the effectiveness of the integrated approach for hippocampus segmentation.

Conclusions:

• The novel method offers a significant advancement in medical image segmentation.
• This approach provides a more accurate and robust solution for segmenting the hippocampus in MR images.