Explicit shape knowledge-based feature augmentation for disease progression analysis: application in liver fibrosis prediction

2020 Fiscal Year Final Research Report

• Project/Area Number: 19K20711
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 90130:Medical systems-related
• Research Institution: Nara Institute of Science and Technology
• Principal Investigator: SOUFI MAZEN 奈良先端科学技術大学院大学, 先端科学技術研究科, 助教 (80823525)
• Project Period (FY): 2019-04-01 – 2021-03-31
• Keywords: 肝線維化 / 統計形状モデリング / MR画像

Outline of Final Research Achievements

In this research, I aimed at the development of disease progression model for prediction of the fibrosis stage in patients with chronic liver diseases. During the research period, I proposed an automated diagnosis approach based on MR images. By using a statistical shape model based on Partial Least Squares (PLS) method, an accuracy for early detection of liver fibrosis of 90±3% was obtained. The developed model did not only represent commonly observed generic variations associated with liver fibrosis, but also localized variations. This has shown the scientific significance of the developed model (IJCARS; Impact Factor: 2.473, JAMIT2019). In addition, a large-scale database of 251 MR images with ground-truth labels for the liver and spleen, and automatic segmentation research was performed (JAMIT2020).

Free Research Field

人間医工学

Academic Significance and Societal Importance of the Research Achievements

本研究では、MR画像における統計形状解析とテクスチャ解析に基づいた特徴量を融合すると、肝線維化ステージの自動診断精度向上を示した。また、統計形状モデルにより、一般的に観察される左葉の全体的な肥大と右葉の収縮のみならず、局所的な変形、例えば右葉後部と尾状葉の肥大も表現できて、統計形状モデルの学術的な意義を確認できた。
さらに、251症例の肝臓と脾臓の正解データを作成し、深層学習に基づいた画像自動セグメンテーションツール(Bayesian U-Net)をMR画像に適応した。予測確信度とセグメンテーション精度(Dice係数)との高い相関を得て、本手法の汎用性を示した。