Deep Separable Spatiotemporal Learning for Fast Dynamic Cardiac MRI.

Area of Science:

• Medical Imaging
• Machine Learning
• Cardiovascular Diagnostics

Background:

• Dynamic magnetic resonance imaging (MRI) is crucial for cardiac diagnosis.
• Undersampling k-space data enables faster MRI but complicates image reconstruction.
• Deep learning reconstruction methods require extensive training data.

Purpose of the Study:

• To propose a novel and efficient deep learning approach for cardiac MRI reconstruction.
• To address the challenge of high-dimensional processing with limited training data.

Main Methods:

• Developed Deep Separable Spatiotemporal Learning (DeepSSL) network.
• Incorporated spatiotemporal priors, temporal low-rankness, and spatial sparsity.
• Unrolled a 2D spatiotemporal reconstruction model iteration process.

Main Results:

• DeepSSL surpasses state-of-the-art methods visually and quantitatively.
• Reduced training data requirements by up to 75%.
• Demonstrated adaptability to new patients and prospectively undersampled real-time MRI, enhancing cardiac segmentation accuracy.

Conclusions:

• DeepSSL is efficient and adaptive, even with highly limited training data.
• Shows promise for high-dimensional data reconstruction in MRI applications.
• Enhances diagnostic capabilities in dynamic cardiac MRI.