[Medical image segmentation based on guided filtering and multi-atlas].

Area of Science:

• Medical Imaging
• Computer Vision
• Neuroimaging

Background:

• Accurate segmentation of anatomical structures in medical images is crucial for diagnosis and treatment planning.
• Existing automatic segmentation methods often struggle with accuracy, smoothness, and robustness.

Purpose of the Study:

• To propose a novel automatic medical image segmentation strategy.
• To enhance the accuracy, smoothness, robustness, and reliability of image segmentation, particularly for brain MRI.

Main Methods:

• A multi-atlas registration framework incorporating guided filtering and thresholding approaches.
• Utilizes atlas prior information for registration and local image information to correct errors.
• Employs label fusion weighted by registration similarity measures.

Main Results:

• Achieved a median Dice coefficient of 86% for left hippocampus and 87.4% for right hippocampus segmentation on brain MRI.
• Demonstrated superior performance compared to traditional label fusion algorithms.
• Showcased good efficiency and accuracy in segmenting the hippocampus region.

Conclusions:

• The proposed guided filtering and multi-atlas based segmentation strategy offers accurate and reliable results for medical image analysis.
• This method represents an advancement over common brain image segmentation techniques.
• The framework effectively segments complex anatomical regions like the hippocampus.