Longitudinal Strain by Artificial Intelligence-Driven Automated Strain Analysis for Left Ventricular Function...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Echocardiography for acute myocardial infarction (AMI) faces challenges in accurately assessing left ventricular (LV) dysfunction and infarcts due to subjective analysis.
• Standardized, quantitative methods are needed to overcome variability in conventional echocardiographic assessments.

Purpose of the Study:

• To validate an artificial intelligence (AI)-driven automated strain framework for standardized quantification of myocardial deformation in AMI.
• To assess the correlation of AI-derived strain parameters with clinical biomarkers of myocardial injury.

Main Methods:

• A retrospective analysis of 102 AMI patients and 90 controls using 2D speckle-tracking echocardiography.
• A ResNet-18 AI model was employed for global/regional longitudinal strain (LPSS) quantification and infarct localization via dual-task learning.
• Training involved two-phase optimization with real-time data augmentation; statistical analyses correlated strain with LV ejection fraction (LVEF), cardiac troponin T (cTnT), and ST-segment elevation.

Main Results:

• AI-derived global LPSS strongly correlated with LVEF (r = -0.609, p < 0.001), outperforming the wall motion score index.
• Infarct-zone LPSS showed the strongest correlation with cTnT (r = 0.671, p < 0.001) and ST elevation (r = 0.321, p = 0.001).
• Reproducibility was confirmed by Bland-Altman analysis (intra-observer bias: 0.7% ± 1.2%; interobserver: 1.1% ± 3.1%).

Conclusions:

• AI-driven strain analysis offers a standardized approach to LV functional assessment in AMI.
• The AI framework provides quantitative correlations with enzymatic and electrophysiological markers of injury.
• Infarct localization and detection of compensatory mechanisms by AI support clinical decision-making in AMI management.