SegMotion-Net: Segmentation-guided motion analysis for early myocardial infarction detection from echocardiography...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Myocardial infarction (MI) detection via echocardiography faces challenges due to subjective interpretation and inter-observer variability.
• Early MI detection is critical to prevent irreversible myocardial damage, highlighting the need for objective and reliable methods.

Purpose of the Study:

• To introduce SegMotion-Net, an interpretable AI framework for enhanced myocardial infarction detection.
• To integrate segmentation-derived anatomical information with motion representation learning for improved diagnostic accuracy.

Main Methods:

• SegMotion-Net employs a task-specific memory mechanism for spatiotemporal feature aggregation across cardiac cycles.
• A memory enhancement module refines representations using predictive masks to mitigate error accumulation.
• An LV wall motion dynamics analysis module captures temporally coherent, region-specific motion patterns indicative of MI.

Main Results:

• SegMotion-Net achieved 93.5% Dice coefficient for left ventricular (LV) wall segmentation on the HMC-QU dataset.
• The framework attained an AUC of 86.7% and an F1 score of 87.5% for MI classification.
• External validation demonstrated cross-center robustness, with performance comparable to experienced cardiologists.

Conclusions:

• SegMotion-Net offers an interpretable and clinically meaningful decision support system for MI detection.
• The framework's explicit modeling of LV wall segmentation and motion dynamics enhances diagnostic reliability.
• This AI-driven approach addresses limitations of subjective echocardiography interpretation in MI diagnosis.