Development of high-resolution fMRI image processing using deep learning for non-invasive and low-cost high-precision brain functional imaging

2023 Fiscal Year Final Research Report

• Project/Area Number: 19K17216
• 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: Ota Junko 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 (90825001)
• Project Period (FY): 2019-04-01 – 2024-03-31
• Keywords: fMRI / 深層学習 / 超解像

Outline of Final Research Achievements

Functional magnetic resonance imaging (fMRI) is utilized as a tool for visualizing brain function, but its spatial resolution is relatively low compared to structural MRI. Achieving high spatial resolution fMRI imaging is challenging due to constraints on temporal resolution. On the other hands, while methods such as deep learning can be used to improve spatial resolution post-imaging, it is difficult to prepare high-resolution ideal fMRI images that serve as the teaching material for learning. In this study, we proposed a new high-resolution fMRI approach by leveraging T2*-weighted images, which are acquired similarly to fMRI through echo-planar imaging, and conducting super-resolution processing using adversarial generative networks trained on T2*-weighted images, exploiting their similar image contrast.

Free Research Field

放射線科学

Academic Significance and Societal Importance of the Research Achievements

本研究は、人工知能を用いた超解像処理により画質が向上するだけでなく、さらに脳機能を局在的に評価できるかどうかを明らかにする点で学術的意義がある。また、新たな撮像装置を導入することなく、脳機能をより精密に評価できるため、汎用性が高く、社会的波及効果が期待できる。