Early Detection of Left Ventricular Dysfunction With Machine Learning-Based Strain Imaging in Aortic Stenosis...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Aortic stenosis (AS) necessitates early detection of left ventricular (LV) dysfunction for effective management.
• Traditional echocardiographic measures like ejection fraction (EF) have limited sensitivity for subtle LV functional changes.
• Strain imaging, often limited to global longitudinal strain (GLS), faces robustness challenges.

Purpose of the Study:

• To introduce a novel, fully automatic algorithm for enhanced detection of LV dysfunction in AS patients.
• To utilize multiple strain imaging parameters for improved diagnostic accuracy.
• To address the limitations of current echocardiographic assessments in AS.

Main Methods:

• Application of supervised machine-learning techniques.
• Classification of data from severe AS patients, chest pain subjects, and healthy volunteers.
• Utilizing a dataset comprising 82 severe AS patients, 96 chest pain subjects, and 319 healthy controls.

Main Results:

• The novel algorithm demonstrated superior performance compared to EF and GLS in differentiating AS patients from healthy volunteers (AUC = 0.97).
• The model significantly outperformed EF and GLS in distinguishing AS patients from chest pain subjects (AUC = 0.95).
• Achieved high diagnostic accuracy, surpassing conventional methods in key patient groups.

Conclusions:

• The developed model offers enhanced diagnostic accuracy for LV dysfunction in AS.
• It provides a clinically interpretable tool leveraging strain imaging potential.
• The algorithm can guide clinical decision-making and improve patient management in AS.