Development of high-resolution multiparametric ASL using deep learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K15802
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: Kyoto College of Medical Science
• Principal Investigator: Ishida Shota 京都医療科学大学, 医療科学部, 助教 (50817559)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: MRI / Arterial spin labeling / Deep learning / Perfusion imaging

Outline of Final Research Achievements

Arterial Spin Labeling (ASL)-MRI is a non-invasive perfusion imaging technique that uses magnetically labeled spins in arterial blood as endogenous diffusible tracers. This technique is used to evaluate cerebral blood flow and metabolism. However, its low spatiotemporal resolution and signal-to-noise ratio cause a decrease in the quantitativity of ASL-derived parameters. This study aimed to address these technical issues associated with ASL by employing deep learning (DL). We have developed a DL-based parameter estimation technique with high accurady and noise immunity. The improvement in accuracy and noise immunity of the quantitative ASL-derived parameters was demonstrated through Monte Carlo simulations, tests on healthy subjects, and patients with Moyamoya disease.

Free Research Field

磁気共鳴医学

Academic Significance and Societal Importance of the Research Achievements

従来の脳循環代謝評価は，放射線被曝による侵襲性と実施可能施設が限られる問題があるため，一般利用可能なASL-MRIによる非侵襲脳循環代謝評価法の確立が急務である．しかし，ASLの様々な技術的問題点のために，現在のところ，従来の侵襲検査法を置き換えることはできていない．ASLの技術的問題点を解決する深層学習ネットワークを開発し，モンテカルロシミュレーション・健常被験者・もやもや病患者においてその臨床的有用性を実証した本研究は，ASL-MRIによる完全非侵襲脳循環代謝評価法確立の端緒として意義深い．