A Feasibility Study of AI-driven Imaging for Ultra-Fast MRI

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K17250
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: National Institutes for Quantum Science and Technology
• Principal Investigator: Umehara Kensuke 国立研究開発法人量子科学技術研究開発機構, QST病院, 主任研究員 (90825077)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 超解像 / 敵対的生成ネットワーク / MRI / 高速撮像 / AIイメージング

Outline of Final Research Achievements

This study demonstrated the feasibility of using a proposed method that applies the super-resolution technique utilizing a generative adversarial network to accelerate MRI scans. By performing post-processing on images acquired in a shorter scan time using existing MRI devices, the proposed method reconstructed high-quality and high-resolution images while maintaining image quality. A comparison with the widely used ultra-fast imaging technique, compressed sensing, indicated the significant usefulness of the proposed method. The proposed method has the potential to provide an effective approach for reducing MRI scan time while maintaining high-quality imaging.

Free Research Field

放射線医学

Academic Significance and Societal Importance of the Research Achievements

画質改善を目的とした従来のAIイメージング研究は，畳み込みニューラルネットワークを応用した研究が多くを占めていた．本研究課題では，敵対的生成ネットワークに焦点を当て，適切なモデル選択と学習により，畳み込みニューラルネットワークを超える画質改善が可能であることを示した．本研究成果により，既存装置を用いたAI画像処理による新たなMRI高速撮像実現の可能性が示唆された．