Content-based image retrieval of 3D brain MRI images focusing on disease characteristics for diagnostic support

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K12656
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 90110:Biomedical engineering-related
• Research Institution: Hosei University
• Principal Investigator: Iyatomi Hitoshi 法政大学, 理工学部, 教授 (10386336)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 類似症例検索 / 脳MRI / 診断支援 / 機械学習 / 低次元表現

Outline of Final Research Achievements

In order to develop a similar case retrieval (CBIR) technology focusing on disease features in 3D brain MR images for the purpose of assisting diagnosis, (1) the development of a high-precision skull stripping technique independent of variations in imaging conditions and individual differences. and (2) the acquisition of an excellent low-dimensional representation of brain MR images for CBIR were carried out in close collaboration with Johns Hopkins University, and were able to achieve results beyond our initial expectations.
For (1), we developed a technology with an original subject posture correction mechanism that achieves the world's highest level of speed and accuracy with a small amount of training data. For (2), a number of results were obtained, including harmonization techniques for data acquired at multiple sites and techniques for acquiring low-dimensional representations with high interpretability while retaining disease characteristics.

Free Research Field

機械学習

Academic Significance and Societal Importance of the Research Achievements

脳MR画像の類似画像検索(CBIR)技術は、先端の医療現場から求められている技術であり、単なる診断支援にとどまらず、医師にまれな病気の気づきを促す意義深い研究である。
またこの技術は他の画像診断技術にも転用可能である。本研究は多くの要素技術から構成され、病徴データを保持しながらのデータの低次元化、注目部位の抽出や分割、多拠点データの調和、モデルの説明性の獲得などは、昨今の機械学習技術の本質そのものである。本技術実現の過程で得られる技術は、幅広いAI技術の進歩にも貢献する。