Development of a new MRI image reconstruction by the fusion of compressed sensing and deep learning

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K19917
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 90:Biomedical engineering and related fields
• Research Institution: Kyoto University
• Principal Investigator: Fujimoto Koji 京都大学, 医学研究科, 特定准教授 (10580110)
• Project Period (FY): 2018-06-29 – 2021-03-31
• Keywords: 圧縮センシング / 深層学習

Outline of Final Research Achievements

By using a measured data from radial scan MRI, simulated data with varying degree of time delay in the acquisition was generated. A correction method for this delay was reported at the international conference （international society for magnetic resonance in medicine, ISMRM 2020, abstract number #0657). For the deep learning network, a network consists of VGG19 and the following upsampling layers (Conv2DTranspose, Dropout, concatenate,　BatchNormalization, Activation) was constructed and trained with 5000 knee MRI images, with a successful image recovery.

Free Research Field

MRI

Academic Significance and Societal Importance of the Research Achievements

画像再構成に深層学習の手法を融合させることができれば、画像再構成に要する時間の飛躍的な短縮がはかれ、臨床応用を加速する大きなポイントになりうる。深層学習を用いたMRI画像再構成研究はまだ萌芽期であり、ネットワークの構成、学習方法の最適化、画質評価の手法など今後も検討が必要である。画質劣化を伴わずにMRIの高速撮像が可能となれば、高齢者や小児など、臨床において長時間のMRI撮像が負担となる患者にとって、大きなメリットとなりうる。