Development and clinical application of a deep-learning model to predict hemodynamic parameters from chest radiographs

2023 Fiscal Year Final Research Report

• Project/Area Number: 20KK0375
• Research Category: Fund for the Promotion of Joint International Research (Fostering Joint International Research (A))
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 53020:Cardiology-related
• Research Institution: Mie University
• Principal Investigator: Toba Shuhei 三重大学, 医学部附属病院, 助教 (20806111)
• Project Period (FY): 2021 – 2023
• Keywords: 胸部レントゲン写真 / 小児循環器学 / 先天性心疾患 / 人工知能 / フォンタン手術 / 国際共同研究

Outline of Final Research Achievements

To improve the performance of artificial intelligence in quantitatively predicting hemodynamic parameters from chest radiograms, we developed a grayscale image-specific deep learning-based model pre-trained using large public datasets, and compared its performance with existing models. The comparison used a public dataset related to pediatric pneumonia diagnosis. In terms of both training time and diagnostic capability, the newly developed model showed improvements over the existing models.

Additionally, using data from Boston Children's Hospital, we developed a deep learning-based model to predict the condition of pulmonary vessels from chest radiograms in preoperative patients undergoing Fontan operation, demonstrating the ability of an artificial intelligence model for quantitative prediction.

Free Research Field

小児循環器学

Academic Significance and Societal Importance of the Research Achievements

本研究の成果により、胸部X線写真から血行動態指標を定量評価するための高性能な人工知能の基盤モデルが得られた。本モデルを応用することで、胸部X線写真からの低侵襲かつ安価な血行動態指標評価手法の開発が加速し、循環器診療における新たな検査手法の確立が期待される。
また胸部X線写真からFontan型手術の予後を予測することが可能となり、今後より正確な手術適応の判断が可能になり、先天性心疾患手術の治療成績が向上することが期待される。