End‐to‐end deep learning nonrigid motion‐corrected reconstruction for highly accelerated free‐breathing coronary MRA

Abstract

To develop an end-to-end deep learning technique for nonrigid motioncorrected (MoCo) reconstruction of ninefold undersampled free-breathing wholeheart coronary MRA (CMRA). Methods: A novel deep learning framework was developed consisting of a diffeomorphic registration network and a motion-informed model-based deep learning (MoDL) reconstruction network. The registration network receives as input highly undersampled (~22×) respiratory-resolved images and outputs 3D nonrigid respiratory motion fields between the images. The motion-informed MoDL performs MoCo reconstruction from undersampled data using the predicted motion fields. The whole deep learning framework, termed as MoCo-MoDL, was trained end-to-end in a supervised manner for simultaneous 3D nonrigid motion estimation and MoCo reconstruction. MoCo-MoDL was compared with a state-of-the-art nonrigid MoCo CMRA reconstruction technique in 15 retrospectively undersampled datasets and 9 prospectively undersampled acquisitions. Results: The acquisition time for ninefold accelerated CMRA was ~2.5 min. The reconstruction time was ~22 s for the proposed MoCo-MoDL and ~35 min for the conventional approach. MoCo-MoDL achieved higher peak SNR (27.86 ± 3.00 vs.26.71 ± 2.79; P < .05) and structural similarity (0.78 ± 0.06 vs. 0.75 ± 0.06; P < .05) than the conventional approach. Similar vessel length and visual image quality score were obtained with the 2 methods, whereas improved vessel sharpness was observed with MoCo-MoDL. Conclusion:An end-to-end deep learning approach was introduced for simultaneous nonrigid motion estimation and MoCo reconstruction of highly undersampled freebreathing whole-heart CMRA. The rapid free-breathing CMRA acquisition together

End-to-End Deep Learning of Non-rigid Groupwise Registration and Reconstruction of Dynamic MRI

“…The registration loss and reconstruction loss are designed for each unrolled iteration to train the proposed GRDRN. For registration, previous works ( 32 , 33 ) have demonstrated that by constructing the MSE loss based on the fully sampled ground truth images, it is possible to learn the motion from undersampled images. Therefore, while the input to GRN is the undersampled or intermediate reconstructed images, the registration loss as defined in Equation (1) for the k -th iteration is calculated based on the fully sampled ground truth dynamic images .…”

“…The motion-augmented dynamic sequence is then incorporated into the reconstruction network to improve the reconstruction performance. For GRN, we employ the self-supervised deep learning registration model, which is more efficient and robust than traditional motion estimation algorithms in the presence of undersampling artifacts ( 32 , 33 ). To the best of our knowledge, this is the first work that embeds groupwise registration network into the deep learning reconstruction framework to exploit the full temporal information of the acquired data to aid in the dynamic MRI reconstruction.…”

Physics-Driven Deep Learning for Computational Magnetic Resonance Imaging

“…The estimated motion fields are then used to enhance the data consistency and exploit the information of all motion resolved states to reconstruct images of the body trunk. In the context of coronary MRI, a motion-informed MoDL network was proposed [11], using diffeomorphic motion fields estimated from the zero-filled images using a UNet and subsequent scaling and squaring layer. These motion fields are then embedded into the data consistency layer, solved via the conjugate gradient algorithm as in MoDL.…”

“…Recently, deep learning methods have emerged as a powerful tool for solving many inverse problems in computational MRI. Among these, MRI reconstruction for accelerated acquisitions remains the most wellstudied [6][7][8], along with several strategies for quantitative MRI [9], motion [10,11] and other non-linear physical models [12,13]. Out of a plethora of approaches for these problems, physics-driven methods have emerged as the most well-received deep learning techniques by the MRI community due to their incorporation of the MR domain knowledge.…”

Artificial intelligence in cardiac magnetic resonance fingerprinting

“…However, the non-linear nature of the problems to be solved leads to long reconstruction times, in addition to the parameter tuning required for an optimal regularization. Several DL-based alternatives have been proposed to overcome these limitations and enable not only highly accelerated acquisitions in short reconstruction times (32,33,(46)(47)(48)(49)(50)(51), but also a wide range or AI-aided solutions for CMR segmentation (40) and analysis or outcome prediction (52,53) among others.…”