Fully automated deep learning MAPSE: retrospective analysis and real-time clinical application

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Mitral annular plane systolic excursion (MAPSE) is a key echocardiographic indicator of left ventricular (LV) function.
• Manual MAPSE measurements are operator-dependent and time-consuming, limiting clinical utility.

Purpose of the Study:

• To develop and validate a deep learning (DL) method for fully automated MAPSE estimation.
• To assess the agreement, reproducibility, time efficiency, and feasibility of the DL method compared to manual measurements.

Main Methods:

• A multistep deep learning model was developed for automated MAPSE estimation.
• The DL method was evaluated in retrospective and prospective cohorts, comparing its performance against manual M-mode and cardiac magnetic resonance (CMR) imaging.

Main Results:

• The DL-MAPSE showed good agreement with manual measurements (bias 2.9 mm) and CMR-derived MAPSE (bias 1.0 mm).
• Real-time DL measurements significantly reduced analysis time by 51% and demonstrated excellent feasibility (96%).
• Both DL and manual methods exhibited good test-retest reproducibility.

Conclusions:

• The novel DL method for automated MAPSE offers excellent feasibility, reproducibility, and agreement with established methods.
• Automated DL-MAPSE has the potential to significantly reduce analysis time and enhance the clinical value of LV systolic function assessment.