Ultrasonic Texture Analysis for Predicting Acute Myocardial Infarction.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Acute myocardial infarction (MI) alters cardiomyocyte structure, affecting myocardial acoustic properties.
• Detecting these changes is crucial for accurate diagnosis and patient management.

Purpose of the Study:

• To evaluate ultrasomics, a cardiac ultrasound-based radiomics technique, for identifying ultrasonic texture and morphologic changes in infarcted myocardium.
• To assess the diagnostic performance of machine learning (ML) models trained on ultrasomics features.

Main Methods:

• Utilized multisource data from 684 participants, including retrospective, prospective, and open-source datasets.
• Extracted handcrafted and deep transfer learning-based ultrasomics features from 2- and 4-chamber echocardiographic views.
• Trained ML models using leave-one-source-out cross-validation for robust external validation.

Main Results:

• The ML model achieved a higher AUC (0.87) than deep transfer learning (0.74) in identifying MI.
• ML probability independently predicted MI, even after adjusting for conventional echocardiographic parameters.
• ML probability demonstrated diagnostic value in differentiating acute MI, even with myocardial dysfunction, and improved prediction when combined with longitudinal strain.

Conclusions:

• Cardiac ultrasomics shows significant potential for distinguishing healthy from infarcted myocardium.
• Further validation in diverse populations is needed to establish its role and incremental value in myocardial tissue characterization.
• Ultrasomics offers a promising advancement beyond conventional echocardiography for myocardial tissue analysis.