Distilling sub-space structure across views for cardiac indices estimation.

Area of Science:

• Medical Imaging Analysis
• Cardiovascular Imaging
• Machine Learning in Healthcare

Background:

• Cardiac indices estimation from multi-view images is crucial for cardiac function assessment.
• Current methods often suffer from low data utilization due to view-specific training.

Purpose of the Study:

• To develop a novel cross-view learning framework for enhanced cardiac indices estimation.
• To fully explore multi-view data for improved cardiac function assessment.

Main Methods:

• Proposed distilling sub-space structure across views using a covariance matrix.
• Employed an alternating convex search algorithm for cross-view learning optimization.
• Trained models with sub-space structure regularization and updated structure based on learned parameters.

Main Results:

• The framework was tested on short axis, 2-chamber, and 4-chamber views using MRI and CT data.
• Demonstrated superior performance in cardiac indices estimation compared to state-of-the-art methods.
• Effectiveness verified through comprehensive experimental evaluations.

Conclusions:

• The proposed cross-view learning framework effectively improves cardiac indices estimation.
• The method enhances data utilization by leveraging sub-space structures across different views.
• Offers a promising approach for accurate cardiac function assessment using multi-view imaging.