Echo-SyncNet: Self-Supervised Cardiac View Synchronization in Echocardiography.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Conventional echocardiography (echo) relies on electrocardiograms (ECGs) for temporal alignment of cardiac views.
• ECG acquisition is often infeasible in emergency or point-of-care settings, necessitating alternative synchronization methods.

Purpose of the Study:

• To introduce Echo-SyncNet, a self-supervised learning framework for synchronizing 2D echo series without human supervision or external inputs.
• To develop a method for temporal alignment of cardiac ultrasound data that overcomes the limitations of ECG dependency.

Main Methods:

• Echo-SyncNet utilizes combined intra-view (temporal ordering and spatial similarity) and inter-view (cross-view dependencies) self-supervision signals.
• A feature-rich, low-dimensional embedding space is learned to enable temporal synchronization of multiple echo cines.
• The framework does not require prior assumptions about specific cardiac views used during training.

Main Results:

• Echo-SyncNet demonstrated successful synchronization of perpendicular cardiac views (Apical 2 and Apical 4 chamber) in 998 patients.
• Learned representations outperformed a supervised deep learning method for cardiac phase detection in 3070 patients.
• The framework achieved accurate one-shot learning for cardiac key-frame detection in 1188 validation studies without fine-tuning.
• Generalization to unseen cardiac views was demonstrated.

Conclusions:

• Echo-SyncNet provides a robust, self-supervised solution for temporal synchronization of echocardiographic views.
• The framework enhances the utility of echocardiography in resource-limited settings and facilitates downstream tasks like key-frame detection.
• This approach advances automated cardiac analysis by removing reliance on ECGs.