Fully Automatic Myocardial Segmentation of Contrast Echocardiography Sequence Using Random Forests Guided by Shape...

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Image Segmentation

Background:

• Myocardial contrast echocardiography (MCE) is crucial for assessing left ventricle function and perfusion in coronary artery disease detection.
• Accurate myocardial segmentation is vital for MCE perfusion quantification but is challenging due to noisy, time-varying images.
• Traditional random forests (RF) for segmentation struggle with contextual information and intensity variations.

Purpose of the Study:

• To develop a fully automatic segmentation pipeline for myocardial segmentation in 2-D MCE data.
• To overcome the limitations of classic RF by incorporating shape prior information.
• To enhance the accuracy and robustness of MCE image analysis.

Main Methods:

• A novel pipeline integrating a statistical shape model (SM) with random forests (RF) for myocardial segmentation.
• Incorporation of a shape model feature to improve RF probability maps.
• Refinement of segmentation using SM fitting to probability maps and a bounding box detection preprocessing step.
• Extension to 2-D+t sequences for temporal consistency.

Main Results:

• The proposed method significantly improves segmentation accuracy on clinical MCE datasets.
• The integration of shape models enhances RF performance by providing contextual and shape prior information.
• The pipeline outperforms existing state-of-the-art methods, including classic RF, active shape models, and image registration.

Conclusions:

• The developed automatic segmentation pipeline effectively addresses limitations of classic RF for MCE analysis.
• The novel approach using shape models enhances myocardial segmentation accuracy and robustness.
• This method offers a significant advancement for automated MCE perfusion quantification and coronary artery disease detection.