2019/10/24 by Mario O. Malavé, Corey A. Baron, Malavé, Mario O. +11
Medicine · #Advanced MRI Techniques and Applications #Cardiac Imaging and Diagnostics #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Medical Imaging Techniques and Applications #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1910.11414
openalex publication_date 2019/10/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Purpose: To rapidly reconstruct undersampled 3D non-Cartesian image-based\nnavigators (iNAVs) using an unrolled deep learning (DL) model for non-rigid\nmotion correction in coronary magnetic resonance angiography (CMRA).\n Methods: An unrolled network is trained to reconstruct beat-to-beat 3D iNAVs\nacquired as part of a CMRA sequence. The unrolled model incorporates a\nnon-uniform FFT operator to perform the data consistency operation, and the\nregularization term is learned by a convolutional neural network (CNN) based on\nthe proximal gradient descent algorithm. The training set includes 6,000 3D\niNAVs acquired from 7 different subjects and 11 scans using a variable-density\n(VD) cones trajectory. For testing, 3D iNAVs from 4 additional subjects are\nreconstructed using the unrolled model. To validate reconstruction accuracy,\nglobal and localized motion estimates from DL model-based 3D iNAVs are compared\nwith those extracted from 3D iNAVs reconstructed with \l1-ESPIRiT.\nThen, the high-resolution coronary MRA images motion corrected with\nautofocusing using the \l1-ESPIRiT and DL model-based 3D iNAVs are\nassessed for differences.\n Results: 3D iNAVs reconstructed using the DL model-based approach and\nconventional \l1-ESPIRiT generate similar global and localized\nmotion estimates and provide equivalent coronary image quality. Reconstruction\nwith the unrolled network completes in a fraction of the time compared to CPU\nand GPU implementations of \l1-ESPIRiT (20x and 3x speed\nincreases, respectively).\n Conclusion: We have developed a deep neural network architecture to\nreconstruct undersampled 3D non-Cartesian VD cones iNAVs. Our approach\ndecreases reconstruction time for 3D iNAVs, while preserving the accuracy of\nnon-rigid motion information offered by them for correction.\n