ACCELERATED CORONARY MRI USING 3D SPIRIT-RAKI WITH SPARSITY REGULARIZATION

Area of Science:

• Cardiovascular Imaging
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Coronary Magnetic Resonance Imaging (MRI) is a vital non-invasive tool for diagnosing coronary artery disease.
• Long scan times, necessary for high-resolution imaging amidst cardiac and respiratory motion, limit its clinical utility.
• Machine learning (ML) methods are emerging as powerful tools to accelerate MRI acquisition.

Purpose of the Study:

• To extend the Robust Artificial-neural-network for k-space Interpolation (RAKI) technique for cardiac MRI to arbitrary undersampling patterns.
• To introduce SPIRiT-RAKI, a novel method leveraging coil self-consistency and deep learning for accelerated coronary MRI.
• To evaluate the performance of SPIRiT-RAKI against existing methods like SPIRiT.

Main Methods:

• Development of SPIRiT-RAKI, a scan-specific deep learning approach utilizing convolutional neural networks to enforce coil self-consistency.
• Application of SPIRiT-RAKI to accelerate right coronary MRI scans with various undersampling patterns and acceleration rates.
• Comparative analysis of image reconstructions generated by SPIRiT-RAKI and SPIRiT.

Main Results:

• SPIRiT-RAKI effectively reconstructs coronary MRI images from arbitrarily undersampled k-space data.
• The method demonstrates a significant reduction in residual aliasing and blurring artifacts compared to the standard SPIRiT reconstruction.
• Improved image quality was observed across different undersampling strategies and acceleration factors.

Conclusions:

• SPIRiT-RAKI represents a significant advancement in accelerating coronary MRI acquisition.
• The technique enhances image quality by mitigating artifacts, making it a promising tool for efficient coronary artery disease diagnosis.
• SPIRiT-RAKI's ability to handle arbitrary undersampling broadens its applicability in clinical cardiovascular MRI.