2017/03/03 by Yoseob Han, Jaejun Yoo, Han, Yo Seob +3 · 1 citation
Engineering · Medicine · #Advanced MRI Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Medical Imaging Techniques and Applications #Photoacoustic and Ultrasonic Imaging
paper · pdf · doi:10.48550/arxiv.1703.01135
openalex publication_date 2017/03/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Purpose: The radial k-space trajectory is a well-established sampling\ntrajectory used in conjunction with magnetic resonance imaging. However, the\nradial k-space trajectory requires a large number of radial lines for\nhigh-resolution reconstruction. Increasing the number of radial lines causes\nlonger acquisition time, making it more difficult for routine clinical use. On\nthe other hand, if we reduce the number of radial lines, streaking artifact\npatterns are unavoidable. To solve this problem, we propose a novel deep\nlearning approach with domain adaptation to restore high-resolution MR images\nfrom under-sampled k-space data.\n Methods: The proposed deep network removes the streaking artifacts from the\nartifact corrupted images. To address the situation given the limited available\ndata, we propose a domain adaptation scheme that employs a pre-trained network\nusing a large number of x-ray computed tomography (CT) or synthesized radial MR\ndatasets, which is then fine-tuned with only a few radial MR datasets.\n Results: The proposed method outperforms existing compressed sensing\nalgorithms, such as the total variation and PR-FOCUSS methods. In addition, the\ncalculation time is several orders of magnitude faster than the total variation\nand PR-FOCUSS methods.Moreover, we found that pre-training using CT or MR data\nfrom similar organ data is more important than pre-training using data from the\nsame modality for different organ.\n Conclusion: We demonstrate the possibility of a domain-adaptation when only a\nlimited amount of MR data is available. The proposed method surpasses the\nexisting compressed sensing algorithms in terms of the image quality and\ncomputation time.\n