Bidirectional segmentation of three-dimensional cardiac MR images using a subject-specific dynamical model

Area of Science:

• Medical image analysis
• Cardiovascular imaging
• Computational anatomy

Background:

• Statistical models are crucial for left ventricle segmentation in cardiac imaging.
• Existing static and generic dynamical models have limitations in capturing individual cardiac motion and temporal dynamics.

Purpose of the Study:

• To develop a subject-specific dynamical model (SSDM) for accurate left ventricle segmentation.
• To address limitations of static and generic models by incorporating inter-subject variability and intra-subject temporal dynamics.

Main Methods:

• Proposed a subject-specific dynamical model (SSDM) integrating inter-subject variability and intra-subject temporal dynamics.
• Developed a dynamic prediction algorithm for progressive shape prediction in cardiac sequences.
• Implemented bidirectional segmentation to mitigate error accumulation by leveraging the periodic nature of cardiac motion.

Main Results:

• The SSDM effectively captures local shape variations in cardiac sequences.
• Bidirectional segmentation significantly suppresses the propagation of segmentation errors.
• Leave-one-out validation on 32 sequences demonstrated the algorithm's efficacy.

Conclusions:

• The proposed SSDM offers a robust approach for left ventricle segmentation.
• Simultaneously handling individual variability and temporal dynamics enhances segmentation accuracy.
• The bidirectional segmentation strategy effectively minimizes cumulative errors in dynamic cardiac imaging.