Tracking-based mitral annular plane systolic excursion (MAPSE) measurement using deep learning in B-mode ultrasound

Mitral annular plane systolic excursion (MAPSE) is an important measure of left ventricular function. Current clinical practice is to measure it manually using M-mode ultrasound imaging which has several disadvantages such as “out-of-line” motion and M-mode angle and operator dependency. In this work, we propose a fully automatic method for measuring MAPSE in B-mode ultrasound using deep learning. The method involves multiple neural networks to detect end-diastolic and end-systolic frames, perform annulus landmark detection, and frame-by-frame tracking. It is also demonstrated how this B-mode based MAPSE can be used to remove radial motion of the annulus from the MAPSE measurement, thereby only measuring longitudinal motion of the annular plane. The landmark detection accuracy in end-diastole was measured to be $3.0\pm 2.5$ mm, while the full pipeline gave a MAPSE accuracy of $-1.5\pm 2.1$ mm on a 72 subject dataset.