Segmentation of B-mode cardiac ultrasound data by Bayesian Probability Maps

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Cardiology

Background:

• Accurate segmentation of cardiac structures like the endocardium is crucial for diagnosing heart conditions.
• Clinical B-mode ultrasound is a widely used imaging modality, but image artifacts can complicate analysis.
• Existing models may not fully address the complexities of endocardial motion and ultrasound signal characteristics.

Purpose of the Study:

• To develop and validate a novel Bayesian model for describing endocardium position distribution in cardiac ultrasound.
• To incorporate spatial, temporal, and shape priors for improved segmentation accuracy.
• To address signal censoring and artifacts inherent in clinical B-mode ultrasound.

Main Methods:

• A Bayesian formulation with priors for spatial/temporal smoothness and preferred shapes/position.
• A shape model considering the endocardium, atrial region, and apex.
• A statistical signal model using a novel censored Gamma mixture model to handle left-censored signals and artifacts.
• Gibbs sampling to estimate the Bayesian Probability Map (BPM) of endocardial pixels.
• Cross-validation for regularization parameter estimation and comparison with existing models.

Main Results:

• The proposed Bayesian model effectively describes endocardium position distribution in ultrasound.
• The novel censored Gamma mixture model successfully handles left-censoring artifacts in B-mode ultrasound.
• The Bayesian Probability Map provides a probabilistic representation of endocardial pixels.
• Model performance was validated through cross-validation and comparison with established methods.

Conclusions:

• The developed Bayesian model offers a robust approach for endocardium segmentation in cardiac ultrasound.
• The method enhances accuracy by integrating prior knowledge and addressing specific ultrasound signal challenges.
• This work contributes a novel tool for quantitative analysis of cardiac function from ultrasound data.