An attempt to discover new insights into spinal disorders using explainable AI and unsupervised learning AI

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K16734
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 56020:Orthopedics-related
• Research Institution: Chiba University
• Principal Investigator: Maki Satoshi 千葉大学, 医学部附属病院, 助教 (00771982)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Keywords: 人工知能 / 脊髄損傷 / 頚部脊髄症

Outline of Final Research Achievements

This study aimed to apply radiomics to predict functional outcomes in spinal cord injury and cervical myelopathy and to demonstrate image features that are easily understandable to humans. We have published findings on the differentiation of spinal infections and Modic changes, prediction of neurological outcomes in spinal cord injury, and segmentation of compressed spinal cord in cervical myelopathy. We also developed a web application for predicting functional outcomes in spinal cord injury and identified important factors. However, in predicting surgical outcomes for OPLL, we found a trade-off between interpretability and predictive accuracy. It became clear that while machine learning models excel in explainability, it is challenging in deep learning for images.

Free Research Field

人口知能／脊髄損傷／頚部脊髄症

Academic Significance and Societal Importance of the Research Achievements

本研究は、脊髄損傷と頚部脊髄症の機能予後予測において、radiomicsを用いることで従来より正確な予測を可能にし、脊髄疾患における有効性を実証した。また、人工知能が着目する画像所見を視覚化することで、客観的な所見に基づく診断を可能にした。これにより、適切な治療選択やリハビリ計画の立案が可能となり、患者のQOL向上が期待できる。さらに、的確な治療方針決定につながり、医療の質の向上にも貢献する。本研究は、radiomicsの有用性を実証し、学術面と臨床面の両面で重要な意義がある。また、radiomicsを他疾患にも応用できれば、幅広い診断・治療の改善が見込まれる。