Real-Time Echocardiography Guidance for Optimized Apical Standard Views

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Cardiac function measurements (e.g., ejection fraction, myocardial strain) rely on 2-D ultrasound.
• Image quality and measurement accuracy are operator-dependent, requiring precise transducer positioning for standard views.

Purpose of the Study:

• To develop and validate a deep learning tool that provides real-time guidance for ultrasound transducer movements.
• To assist users in navigating toward standard echocardiographic views, enhancing ease of use and image standardization.

Main Methods:

• A deep learning model was trained using 3-D ultrasound volumes to simulate 2-D transducer movements.
• Neural networks calculated transducer position in a regression manner.
• The method was validated on 2-D echocardiographic datasets in a clinical setting.

Main Results:

• The tool suggested adequate transducer movements 75% of the time across all degrees of freedom.
• It achieved 95% accuracy for transducer rotation guidance.
• Real-time application demonstrated the correlation between suggested movements, image acquisition, and user feedback.

Conclusions:

• The proposed deep learning tool effectively guides ultrasound transducer positioning for echocardiography.
• This technology can simplify the process for less experienced users and standardize imaging for all skill levels.
• The tool has the potential to improve the reliability and consistency of cardiac function measurements.