Synthetic echocardiographic image sequences for cardiac inverse electro-kinematic learning

Area of Science:

• Computational biology
• Medical imaging
• Machine learning

Background:

• Understanding the relationship between cardiac electrical activity and mechanical motion is crucial for diagnosing heart conditions.
• Current methods for assessing cardiac electromechanical coupling often rely on invasive procedures or limited imaging modalities.

Purpose of the Study:

• To develop a synthetic 3D echocardiography (US) image database using cardiac electromechanical modeling.
• To train a machine learning algorithm for estimating cardiac depolarization times from kinematic descriptors.
• To evaluate the algorithm's performance on synthetic and real clinical data.

Main Methods:

• A cardiac electromechanical model was used to simulate synthetic 3D US image time series.
• A software pipeline involving motion tracking and segmentation was employed to generate synthetic sequences.
• A machine learning algorithm was trained on invariant kinematic descriptors (e.g., local displacements, strains) to estimate depolarization times.
• The algorithm was validated using synthetic data and clinical 3D US sequences from patients with Left Bundle Branch Block.

Main Results:

• A comprehensive synthetic database of 3D US images was successfully created.
• The machine learning algorithm demonstrated the ability to estimate cardiac depolarization times from kinematic features.
• Initial experiments showed promising results in inverse electrokinematic learning on both synthetic and real patient data.

Conclusions:

• Synthetic 3D US data generated from electromechanical models can effectively train machine learning algorithms for cardiac electrokinetics.
• This approach offers a non-invasive method for studying the relationship between electrical disorders and kinematic patterns in the heart.
• The developed method shows potential for improving the diagnosis and understanding of conditions like Left Bundle Branch Block.