Examination of image quality improvement and imaging time reduction of brain nuclear medicine scan using deep learning model

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K16705
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: Yokohama City University
• Principal Investigator: ISHIWATA YOSHINOBU 横浜市立大学, 医学部, 助教 (70835868)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: 核医学 / AI / 人工知能 / 深層学習 / イオフルパン

Outline of Final Research Achievements

In this study, we investigated the reduction of imaging time using artificial intelligence by utilizing 207 cases of 123I-ioflupane previously imaged at our hospital. Original images and images with one-fifth of the imaging time were output from the workstation attached to the imaging equipment, and an artificial intelligence model was constructed. U-Net and its derivative models were employed to build the AI model. The generated AI images and the original images were compared and evaluated using quantitative values (PSNR; Peak Signal to Noise Ratio, SSIM; Structural Similarity) and a reading experiment. As a result, the agreement rate between the quantitative values and the reading experiment results with the original images showed favorable outcomes.

Free Research Field

放射線医学

Academic Significance and Societal Importance of the Research Achievements

123I-ioflupaneは認知症診断に重要な役割を担っており、臨床的・社会的に重要であるが、撮像時間も長く、患者負担も大きい。また、検査効率や経済性の観点からも脳核医学検査の時間短縮は取り組むべき課題と考えられた。本研究では人工知能を用いた画質改善技術を開発し、撮像時間が1/5となる5分間の画像を用いて、オリジナル画像と良好な読影結果の一致率を示すことができた。この結果により、さらなる研究の後、123I-ioflupaneの撮像時間短縮により患者負担の軽減のみならず検査室のスループットの向上にも繋がると考えられた。また、本技術は他の核医学検査にも応用可能であり、その意義は大きいと思われた。