DYNAMIC MRI USING DEEP MANIFOLD SELF-LEARNING.

Area of Science:

• Medical Imaging
• Machine Learning
• Cardiovascular Imaging

Background:

• Cardiac CINE MRI is crucial for assessing heart function.
• Acquiring high-quality cardiac MRI is challenging due to motion and undersampling.
• Existing reconstruction methods may struggle with free-breathing, ungated data.

Purpose of the Study:

• To develop a novel deep self-learning algorithm for cardiac CINE MRI reconstruction.
• To accurately learn the manifold structure of dynamic cardiac data.
• To improve image quality by reducing spatial and temporal blurring.

Main Methods:

• Utilized an autoencoder network to learn the manifold structure from navigator data.
• Integrated the trained autoencoder as a prior within an image reconstruction framework.
• Applied the method to free-breathing, ungated cardiac CINE MRI data acquired with a navigated golden-angle radial sequence.

Main Results:

• The proposed deep self-learning algorithm effectively captured the manifold structure of the cardiac data.
• Reconstructed cardiac CINE MRI images showed reduced spatial and temporal blurring.
• Demonstrated superior performance compared to the SToRM reconstruction method.

Conclusions:

• Deep self-learning offers a powerful approach for cardiac MRI reconstruction from undersampled data.
• The autoencoder-based prior enhances the ability to capture dynamic data manifolds.
• This method holds promise for improving the efficiency and quality of free-breathing cardiac MRI.