Cardiac Valve Event Timing in Echocardiography Using Deep Learning and Triplane Recordings.

Area of Science:

• Medical Imaging
• Cardiovascular Ultrasound
• Artificial Intelligence in Medicine

Background:

• Accurate cardiac valve event timing is vital for echocardiography measurements.
• Current automated methods require external sensors, and manual timing can be inconsistent.
• Existing deep learning models primarily detect only end-diastole (ED) and end-systole (ES).

Purpose of the Study:

• To develop a deep learning approach for enhanced detection of multiple cardiac valve events in echocardiography.
• To leverage triplane recordings for improved accuracy in cardiac timing.
• To overcome limitations of existing automated and manual measurement techniques.

Main Methods:

• A novel deep learning model was developed utilizing triplane echocardiographic recordings.
• The method was trained and validated on data from 240 patients.
• Performance was evaluated using average absolute frame difference (aFD) and cross-validation.

Main Results:

• The deep learning approach accurately detected six cardiac valve events, including those beyond ED and ES.
• Achieved an average absolute frame difference (aFD) as low as 0.6 frames (12 ms) for mitral valve opening.
• Demonstrated robust performance on an independent test set with a worst-case aFD of 1.8 frames (30 ms).

Conclusions:

• The proposed deep learning method significantly enhances the detection of cardiac valve events in echocardiography.
• This approach offers potential for more accurate, rapid, and comprehensive clinical measurements.
• The technique can substantially impact clinical practice by improving diagnostic capabilities.