CardiacField: computational echocardiography for automated heart function estimation using two-dimensional...

Area of Science:

• Computational imaging
• Cardiovascular diagnostics
• Medical artificial intelligence

Background:

• Accurate heart function estimation is crucial for cardiovascular disease management.
• Two-dimensional echocardiography (2DE) has limitations including required expertise and lack of 3D data.
• Existing methods can be complex and prone to variability.

Purpose of the Study:

• To introduce CardiacField, an automated system for precise left ventricular (LV) and right ventricular (RV) ejection fraction (EF) estimation using 2DE.
• To evaluate the usability and performance of CardiacField among novice users and compare it to expert interpretations and deep learning (DL) tools.

Main Methods:

• Developed an implicit neural representation network to reconstruct 3D cardiac volumes from multi-view 2DE images.
• Implemented automatic segmentation for LV and RV areas to calculate EF.
• Assessed accuracy against expert readings and 2D DL models in 127 patients.

Main Results:

• CardiacField processed 2D echocardiograms into 3D heart models in under 2 minutes.
• Achieved a Mean Absolute Error (MAE) of [specific MAE value] for LVEF and [specific MAE value] for RVEF.
• Demonstrated usability among novice users across various ultrasound machines.

Conclusions:

• CardiacField provides accurate EF estimations for LV and RV function comparable to 3D echocardiography.
• The system utilizes a simple apical ring scan with a cost-effective 2DE probe.
• Offers an accessible tool for non-cardiovascular specialists to assess cardiac function.