Gadolinium-Free Cardiac MRI Myocardial Scar Detection by 4D Convolution Factorization

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Gadolinium-based contrast agents are standard for myocardial scar characterization in cardiac magnetic resonance (CMR) imaging.
• Deep learning models show potential for contrast-free scar detection using wall motion abnormalities (WMA) in ischemic patients.
• Limitations exist as WMA can occur without scars, and scars may not present with WMA, especially in non-ischemic heart disease.

Purpose of the Study:

• To develop a novel deep learning model for accurate myocardial scar detection in both ischemic and non-ischemic heart diseases using contrast-free CMR cine images.
• To overcome the limitations of solely relying on wall motion abnormalities for scar detection.

Main Methods:

• A deep spatiotemporal residual attention network (ST-RAN) was developed, integrating factorized 4D convolutional layers and spatiotemporal attention mechanisms.
• The model extracts 3D spatial features and 1D temporal features to capture heart motion and long-range temporal relationships.
• Residual attention blocks were employed to extract multi-scale spatial and temporal features for detecting subtle scar-related changes.

Main Results:

• The ST-RAN model was trained and validated on a large dataset of 3000 patients across different field strengths (1.5T, 3T) and vendors (GE, Siemens), demonstrating generalizability and robustness.
• The model successfully detected myocardial scars in both ischemic and non-ischemic heart disease patients.
• Performance surpassed existing state-of-the-art methods for contrast-free scar detection.

Conclusions:

• The proposed ST-RAN model offers a robust and accurate method for contrast-free myocardial scar detection in CMR.
• This approach holds significant promise for improving the diagnosis and management of various heart conditions.
• The developed model addresses key challenges in scar detection, particularly in non-ischemic cardiomyopathies.