Automatic multi-view pose estimation in focused cardiac ultrasound.

Area of Science:

• Medical Imaging
• Cardiovascular Ultrasound
• Artificial Intelligence in Medicine

Background:

• Focused Cardiac Ultrasound (FoCUS) is a crucial point-of-care tool for cardiovascular assessment.
• Current FoCUS exams are predominantly qualitative 2D due to equipment and operator limitations.
• There is a need for quantitative 3D assessments in FoCUS.

Purpose of the Study:

• To develop a novel framework for automatically estimating the 3D spatial relationships between standard FoCUS views.
• To enable quantitative 3D analysis and improve the diagnostic capabilities of FoCUS.

Main Methods:

• A multi-view U-Net-like fully convolutional neural network was employed.
• The network regressed line-based heatmaps to identify image intersections.
• A system of nonlinear equations was solved to determine the relative 3D pose between FoCUS views.

Main Results:

• The proposed framework successfully estimated the relative 3D poses of FoCUS images.
• Validation on a realistic in silico FoCUS dataset demonstrated promising accuracy.
• Preliminary experiments confirmed the feasibility of 3D image analysis methods.

Conclusions:

• The developed framework enables automatic 3D spatial relationship estimation for FoCUS.
• This approach facilitates the transition from qualitative 2D to quantitative 3D FoCUS assessments.
• The method holds potential for enhancing diagnostic accuracy and clinical utility of FoCUS.