High-dimensional organ state reconstruction during therapy using complex sensing

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K19918
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 90:Biomedical engineering and related fields
• Research Institution: Kyoto University
• Principal Investigator: Nakao Megumi 京都大学, 情報学研究科, 准教授 (10362526)
• Project Period (FY): 2018-06-29 – 2021-03-31
• Keywords: 機械学習 / 臓器変形 / 治療支援 / 医用人工知能

Outline of Final Research Achievements

The purpose of this study is to reconstruct high-dimensional state of patient-specific organs from low-dimensional and local information during treatment by using prior knowledge on organ structures and treatment procedure. A deformable mesh registration method was newly proposed and a statistical deformation model of multiple abdominal organs was constructed. We also built an Image-to-graph convolutional network that estimates the 3D shape from the 2D projection image. We confirmed the estimation accuracy of the 3D shape and deformation of the liver associated with respiration from only a single 2D X-ray image. The shape features of surrounding organs that can accurately estimate the respiratory motion/deformation of pancreatic cancer were identified.

Free Research Field

医用システム

Academic Significance and Societal Importance of the Research Achievements

治療時における生体臓器の幾何学的・力学的状態の理解は生体医工学や放射線治療，コンピュータ外科学において重要なテーマであり，臨床におけるニーズも高い研究課題である．本研究では，治療時の追加計測を前提とせず，計算機で低次元かつ局所的な情報から臓器の形状や変形を再構成可能な枠組みを構築した．統計的変位モデルは生体臓器の3次元形状と変形をより低次元の画像特徴などから推定する目的において事前知識，臓器データベースとして利用可能である． また，2次元X線画像では検出されない膵がんの呼吸性変位を推定可能な周辺臓器の形状特徴量を同定し，臨床利用可能な推定精度を達成することを示した．