Accelerated Non-Contrast-Enhanced Three-Dimensional Cardiovascular Magnetic Resonance Deep Learning Reconstruction

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Cardiovascular magnetic resonance (CMR) is crucial but time-consuming.
• Rapid imaging techniques accelerate acquisition but can reduce image quality.
• The efficacy of Adaptive CS-Net for non-contrast 3D whole-heart REACT imaging is uncertain.

Purpose of the Study:

• To evaluate the effectiveness of Adaptive CS-Net for non-contrast 3D whole-heart REACT imaging.
• To compare image quality and vessel measurements between Adaptive CS-Net reconstructed REACT images, conventional CS reconstructed REACT images, and standard 3D bSSFP sequences.

Main Methods:

• Thirty participants underwent non-contrast modified REACT and standard 3D bSSFP sequences.
• REACT data were acquired with six-fold undersampling and reconstructed using conventional CS and Adaptive CS-Net.
• Image quality and vessel cross-sectional areas were assessed and statistically compared using Friedman and Dunn's post-hoc tests.

Main Results:

• Adaptive CS-Net and CS reconstructed REACT images showed superior image quality for pulmonary veins, neck, and upper thoracic vessels compared to 3D bSSFP.
• Both REACT reconstruction methods yielded significantly higher contrast-to-noise ratio (CNR) for pulmonary veins, ascending aorta, and superior vena cava than 3D bSSFP (p < 0.05).
• Adaptive CS-Net reconstruction demonstrated comparable or superior image quality to conventional CS.

Conclusions:

• Adaptive CS-Net reconstruction for REACT images consistently provides superior or comparable image quality to conventional CS.
• Adaptive CS-Net significantly enhances image quality in pulmonary veins, neck, and upper thoracic vessels compared to standard 3D bSSFP.
• This deep learning approach offers a promising advancement for rapid, high-quality non-contrast cardiovascular magnetic resonance imaging.