의학물리학

본문글자크기
  • [Med Phys.] Reconstruction of multicontrast MR images through deep learning.

    [Med Phys.] Reconstruction of multicontrast MR images through deep learning.

    서울의대, KAIST / 도원준, 서성훈, 최승홍*, 박성홍*

  • 출처
    Med Phys.
  • 등재일
    2020 Mar
  • 저널이슈번호
    47(3):983-997. doi: 10.1002/mp.14006. Epub 2020 Jan 28.
  • 내용

    바로가기  >

    Abstract
    PURPOSE:
    Magnetic resonance (MR) imaging with a long scan time can lead to degraded images due to patient motion, patient discomfort, and increased costs. For these reasons, the role of rapid MR imaging is important. In this study, we propose the joint reconstruction of multicontrast brain MR images from down-sampled data to accelerate the data acquisition process using a novel deep-learning network.

    METHODS:
    Twenty-one healthy volunteers (female/male = 7/14, age = 26 ± 4 yr, range 22-35 yr) and 16 postoperative patients (female/male = 7/9, age = 49 ± 9 yr, range 37-62 yr) were scanned on a 3T whole-body scanner for prospective and retrospective studies, respectively, using both T1-weighted spin-echo (SE) and T2-weighted fast spin-echo (FSE) sequences. We proposed a network which we term "X-net" to reconstruct both T1- and T2-weighted images from down-sampled images as well as a network termed "Y-net" which reconstructs T2-weighted images from highly down-sampled T2-weighted images and fully sampled T1-weighted images. Both X-net and Y-net are composed of two concatenated subnetworks. We investigate optimal sampling patterns, the optimal patch size for augmentation, and the optimal acceleration factors for network training. An additional Y-net combined with a generative adversarial network (GAN) was also implemented and tested to investigate the effects of the GAN on the Y-net performance. Single- and joint-reconstruction parallel-imaging and compressed-sensing algorithms along with a conventional U-net were also tested and compared with the proposed networks. For this comparison, the structural similarity (SSIM), normalized mean square error (NMSE), and Fréchet inception distance (FID) were calculated between the outputs of the networks and fully sampled images. The statistical significance of the performance was evaluated by assessing the interclass correlation and in paired t-tests.

    RESULTS:
    The outputs from the two concatenated subnetworks were closer to the fully sampled images compared to those from one subnetwork, with this result showing statistical significance. Uniform down-sampling led to a statically significant improvement in the image quality compared to random or central down-sampling patterns. In addition, the proposed networks provided higher SSIM and NMSE values than U-net, compressed-sensing, and parallel-imaging algorithms, all at statistically significant levels. The GAN-based Y-net showed a better FID and more realistic images compared to a non-GAN-based Y-net. The performance capabilities of the networks were similar between normal subjects and patients.

    CONCLUSIONS:
    The proposed X-net and Y-net effectively reconstructed full images from down-sampled images, outperforming the conventional parallel-imaging, compressed-sensing and U-net methods and providing more realistic images in combination with a GAN. The developed networks potentially enable us to accelerate multicontrast anatomical MR imaging in routine clinical studies including T1-and T2-weighted imaging.

     

     

     


    Author information

    Do WJ1, Seo S1, Han Y1, Ye JC1, Choi SH2, Park SH1.
    1
    Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea.
    2
    Department of Radiology, Seoul National University College of Medicine, Seoul, Korea.

  • 키워드
    deep learning; multicontrast MRI; x-net; y-net
  • 연구소개
    본 연구팀은 MRI의 다중 대조도 영상을 복원하기 위한 새로운 딥러닝 네트워크를 개발했다. 이번 연구를 통해 병원에서 routine하게 획득하는 다중 대조도 MRI 영상을 얻는 시간이 크게 감소되어 환자의 편이성과 촬영비용 절감 등의 효과를 기대할 수 있을 것으로 보인다. 일반적으로 임상적 환경에서 MRI 촬영은 정확한 진단을 위해 두 개 이상의 대조도로 진행되지만, 이에 따라 촬영 시간 또한 증가된다. 길어진 영상 획득 시간은 비싼 MRI 촬영 비용과도 연관이 있고 환자들의 불편함을 유발할 수 있으며, 영상의 질 또한 환자의 움직임 등으로 인한 인공물에 더 취약해 질 수 있다. 본 연구팀은, 임상에서 보다 정확한 진단을 위해 MRI 영상을 다중 대조도로 얻는다는 점을 활용해 효율을 높여 촬영시간을 최대 8배까지 줄여 영상을 복원하였으며, 실제로 데이터를 얻을 당시의 전략을 고려해 네트워크들을 따로 개발하였다. 구체적으로, (i) 다중 대조도 전체 프로토콜의 촬영시간을 모두 줄이는 네트워크(X-net)와 (ii) 하나의 프로토콜은 전체 인코딩 데이터를 획득하고 나머지 프로토콜들은 촬영시간을 크게 줄이는 네트워크(Y-net)를 따로 개발하여, MRI 다중 대조도 영상을 촬영하는 목적에 맞춰 다르게 최적화하였다. 본 연구는 ‘메디컬 피직스 (Medical Physics)’ 2020년 3월호 표지 논문으로 게재되었다.
  • 덧글달기
    덧글달기
       IP : 18.206.160.129

    등록