ESPNet: An Automated Left Ventricular Ejection Fraction Measurement via Multi-Frame Temporal Aggregation

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Left ventricular ejection fraction (LVEF) is crucial for cardiac function assessment via echocardiography.
• Manual LVEF measurement is time-consuming, subjective, and challenging in complex cases.
• Existing automated methods lack consistency and reliability, especially in heart failure or arrhythmia.

Purpose of the Study:

• To develop an innovative automated approach for LVEF measurement using multi-frame temporal aggregation.
• To enhance the accuracy and efficiency of LVEF assessment in echocardiography.
• To overcome limitations of single-view analysis and improve reliability in challenging conditions.

Main Methods:

• Introduced EchoSegPhaseNet (ESP Net), an automated LVEF measurement system.
• Utilized a two-component architecture: PhaseFormer for cardiac phase detection and LVFormer for endocardial segmentation.
• Employed multi-frame temporal aggregation and spatial-temporal feature fusion from apical four-chamber (A4C) and two-chamber (A2C) views.

Main Results:

• ESP Net demonstrated faster inference speed in phase detection compared to existing methods.
• Achieved superior speed and accuracy in endocardial segmentation, outperforming current automated techniques.
• Reported a Pearson correlation coefficient of 0.981, R2 of 0.96, and MAE of 3.281 against manual measurements.

Conclusions:

• ESP Net provides a reliable and efficient automated solution for LVEF measurement.
• The system achieves consistency comparable to inter-observer variability among clinicians.
• Addresses the clinical need for robust automated LVEF assessment in echocardiography.