Advancing cardiac motion estimation with emerging AI techniques for enhanced echocardiographic image registration

Area of Science:

• Cardiovascular imaging and AI
• Medical image analysis
• Biomedical engineering

Background:

• Cardiac motion estimation is crucial for diagnosing cardiovascular diseases.
• Current echocardiographic image registration methods suffer from low resolution, noise, and anatomical distortion.
• Accurate cardiac motion tracking is essential for effective clinical diagnosis.

Purpose of the Study:

• To develop an AI-powered architecture for enhanced cardiac motion prediction and echocardiographic image registration.
• To improve the accuracy, efficiency, and generalisability of cardiac motion analysis.
• To overcome limitations of existing methods in echocardiography.

Main Methods:

• Utilized a novel AI architecture combining Vision Transformers, Diffusion Models, and Neural Radiance Fields (NeRF).
• Employed adversarial and self-supervised contrastive learning to enhance image quality and generalisability.
• Integrated a graph neural network (GNN)-based anatomical constraint to maintain heart shape integrity.

Main Results:

• Achieved high-quality cardiac motion prediction using the integrated AI models.
• Demonstrated improved echocardiographic registration across adult and foetal datasets through advanced learning techniques.
• Successfully maintained accurate heart morphology during motion analysis via GNN constraints.

Conclusions:

• The proposed AI architecture enables more accurate, efficient, and real-time cardiac motion tracking.
• This approach reduces reliance on large labeled datasets, making advanced analysis more accessible.
• The innovative method enhances echocardiographic image registration, offering therapeutic viability across diverse patient populations.