Acceleration is an important consideration when imaging moving organs such as the heart. Not only does acceleration enable motion-free scans but, more importantly, it lies at the heart of capturing the dynamics of cardiac motion. For over three decades, various ingenious approaches have been devised and implemented for rapid CINE MRI suitable for dynamic cardiac imaging. Virtually all techniques relied on acquiring less data to reduce acquisition times. Parallel imaging was among the first of these innovations, using multiple receiver coils and mathematical algorithms for reconstruction; acceleration factors of 2 to 3 were readily achieved in clinical practice. However, in the context of imaging dynamic events, further decreases in scan time beyond those provided by parallel imaging were possible by exploiting temporal coherencies. This recognition ushered in the era of k-t accelerated MRI, which utilized predominantly statistical methods for image reconstruction from highly undersampled k-space. Despite the successes of k-t acceleration methods, however, the accuracy of reconstruction was not always guaranteed. To address this gap, MR physicists and mathematicians applied compressed sensing theory to ensure reconstruction accuracy. Reconstruction was, indeed, more robust, but it required optimizing regularization parameters and long reconstruction times. To solve the limitations of all previous methods, researchers have turned to artificial intelligence and deep neural networks for the better part of the past decade, with recent results showing rapid, robust reconstruction. This review provides a comprehensive overview of key developments in the history of CINE MRI acceleration, and offers a unique and intuitive explanation behind the techniques and underlying mathematics.Level of Evidence: 5Technical Efficacy Stage: 1.
Keywords: acceleration; artificial intelligence; compressed sensing; deep learning; parallel imaging; real-time imaging.
© 2020 International Society for Magnetic Resonance in Medicine.