Automatic segmentation of a fetal echocardiogram using modified active appearance models and sparse representation

Area of Science:

• Medical Imaging
• Cardiology
• Biomedical Engineering

Background:

• Accurate segmentation of the fetal left ventricle is crucial for prenatal cardiac assessment.
• Traditional segmentation methods face challenges with image noise (speckle) and texture variations in fetal echocardiograms.

Purpose of the Study:

• To develop and evaluate a novel automated approach for segmenting the fetal left ventricle in echocardiograms.
• To improve segmentation accuracy and robustness compared to existing methods.

Main Methods:

• Integration of sparse representation, global constraint, and local refinement within an active appearance model (AAM) framework.
• Development of an enhanced AAM texture model to address speckle and texture ambiguities.
• Utilizing sparse representation for initial pose localization and incorporating globally constrained points and clinically relevant local features for convergence.

Main Results:

• The proposed approach achieved segmentation accuracies of 84.12% (synthetic data) and 84.39% (clinical data) for overlapped area, outperforming traditional ASM, AAM, and globally constrained AAM.
• Sparse representative methods significantly improved initialization.
• The method demonstrated effectiveness in detecting the fetal left ventricle.

Conclusions:

• The novel AAM-based approach with sparse representation provides superior automated segmentation of the fetal left ventricle.
• This technique offers improved accuracy and robustness for prenatal cardiac imaging analysis.