Hybrid method for iterative CT reconstruction and deep learning

• Project/Area Number: 19K08093
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: Hirosaki University (2021-2022); The University of Tokyo (2019-2020)
• Principal Investigator: Ozaki Sho 弘前大学, 理工学研究科, 助教 (60615326)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Project Status: Completed (Fiscal Year 2022)
• Budget Amount: ¥3,510,000 (Direct Cost: ¥2,700,000、Indirect Cost: ¥810,000); Fiscal Year 2021: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000); Fiscal Year 2020: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000); Fiscal Year 2019: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
• Keywords: 画像再構成 / 深層学習 / CT / 画像誘導放射線治療 / 深層ニューラルネット / 画像変換 / モダリティ変換

Outline of Research at the Start

近年、深層学習を用いた超解像技術は飛躍的に進歩している。一方で、情報が不足した疎な投影データからの画像再構成は、これまで逐次近似法がよく用いられて研究されてきた。深層学習による超解像技術はビックデータからの情報によって、不足した情報を補い画質を改善させる方法であるのに対して、逐次近似法は人間の知識によって情報を補い画質を改善させる。本研究課題では、ビックデータに基づく深層学習と人間の知識に基づく逐次近似法を相補的に組み合わせることによって、まったく新しい画像再構成法を開発する。さらにこのハイブリッド画像再構成法を画像誘導放射線治療に応用することによって、より高い精度の放射線治療を実現する。

Outline of Final Research Achievements

We develop a hybrid method of iterative CT reconstruction and deep learning. For the pre-training of generator and discriminator, we train CycleGAN model to improve Mega-voltage CT (MVCT). We introduce several loss functions, which impose structure preservation of the input image. Our deep learning model nicely improve image quality of MVCT with reducing the structure changes. Then, we incorporate the generator into the iterative CT reconstruction. This hybrid CT reconstruction method exceeds the previous reconstruction methods such as iterative reconstruction and deep image prior reconstruction in terms of SSIM and PSNR.

Academic Significance and Societal Importance of the Research Achievements

MVCTやCone-beam CTなどの位置照合用CTは、高精度の画像誘導放射線治療を行う上で必須の装置である。しかしながら一般にこれらのCTの画質は低くく、これは治療の精度に関わる問題である。本研究で開発した深層学習モデルやハイブリッド画像再構成法は、位置照合用CTの画質を大幅に改善した。この成果は、画像誘導放射線治療の際の位置照合の精度を高め、より高精度の放射線治療を実現できると期待される。

Research Products

• [Int'l Joint Research] Accuray Incorporated(米国)
• [Journal Article] Training of deep cross‐modality conversion models with a small dataset, and their application in megavoltage CT to kilovoltage CT conversion (2022)
  • Author(s): Ozaki S, Kaji S, Nawa K, Imae T, Aoki A, Nakamoto T, Ohta T, Nozawa Y, Yamashita H, Haga A, Nakagawa K
  • Journal Title: Medical Physics Volume: - Issue: 6 Pages: 1-14
  • DOI: 10.1002/mp.15626
  • Peer Reviewed / Open Access
• [Journal Article] Training modality conversion models with small data and its application to MVCT to kVCT conversion (2021)
  • Author(s): S. Ozaki, S. Kaji, K. Nawa, T. Imae, A. Aoki, T. Nakamoto, T. Ohta, Y. Nozawa, A. Haga and K. Nakagawa
  • Journal Title: Radiotherapy and Oncology Volume: 161 Pages: 586-587
  • DOI: 10.1016/s0167-8140(21)07034-1
  • Peer Reviewed
• [Journal Article] Fast Statistical Iterative Reconstruction for Mega-voltage Computed Tomography (2020)
  • Author(s): S. Ozaki, A. Haga, E. Chao, C. Maurer, K. Nawa, T. Ohta, T. Nakamoto, Y. Nozawa, T. Magome, M. Nakano, K. Nakagawa
  • Journal Title: The Journal of Medical Investigation Volume: 67 Issue: 1.2 Pages: 30-39
  • DOI: 10.2152/jmi.67.30
  • NAID: 130007839921
  • ISSN: 1343-1420, 1349-6867
  • Peer Reviewed / Open Access / Int'l Joint Research
• [Journal Article] Visual enhancement of Cone‐beam CT by use of CycleGAN (2020)
  • Author(s): Kida S, Kaji S, Nawa K, Imae T, Nakamoto T, Ozaki S, Ohta T, Nozawa Y, Nakagawa K.
  • Journal Title: Medical Physics Volume: 47(3) Issue: 3 Pages: 998-1010
  • DOI: 10.1002/mp.13963
  • Peer Reviewed / Open Access
• [Presentation] Image quality enhancement of medical images by use of deep learning with a small amount of training data (2023)
  • Author(s): Sho Ozaki
  • Organizer: Interdisciplinary Science Conference in Okinawa (ISCO 2023) -Physics and Mathematics meet Medical Science-
  • Int'l Joint Research
• [Presentation] Iterative CT reconstruction with deep neural networks (2023)
  • Author(s): Sho Ozaki
  • Organizer: The 2nd International Conference on Radiological Physics and Technology (ICRTP)
  • Int'l Joint Research
• [Presentation] Iterative reconstruction of MVCT with deep neural networks (2023)
  • Author(s): Sho Ozaki
  • Organizer: ESTRO 2023 Annual Meeting
  • Int'l Joint Research
• [Presentation] Training modality conversion models with small data and its application to MVCT to kVCT conversion (2021)
  • Author(s): Sho Ozaki
  • Organizer: ESTRO 2021 Annual Meeting
  • Int'l Joint Research
• [Presentation] Denoising and Contrast Enhancement of MVCT Using Deep Learning-based Methods (2021)
  • Author(s): 尾崎翔
  • Organizer: 第121回日本医学物理学会学術大会
• [Presentation] Denoising and Contrast Enhancement of MVCT Using Deep Learning-based methods (2021)
  • Author(s): 尾崎翔
  • Organizer: 第121回日本医学物理学会学術集会