人工智能与影像组学用于骨质疏松性骨折诊断与预测的系统综述
本系统综述检索了2021年1月至2025年3月PubMed、Embase和Web of Science的119篇文献,最终纳入20项研究,评估机器学习与影像组学基于CT和MRI进行骨质疏松评估及骨折风险预测的价值。作者认为该技术潜力较大,但未来需更大规模多中心数据集和更广泛的临床整合。
要点
- 影像组学特征联合临床数据可提升骨折风险预测模型的准确性。
- 常用算法包括逻辑回归、极端梯度提升、支持向量机、随机森林和神经网络。
核心结论
- 机器学习和影像组学在骨质疏松及骨折风险评估方面显示出较大潜力。
- 未来研究应侧重更大规模的多中心数据集和更广泛的临床整合,以提升模型稳健性和临床适用性。
研究设计与样本
- 系统综述,检索PubMed、Embase和Web of Science,关键词为radiomics、osteoporosis和fracture。
- 共识别2021年1月15日至2025年3月15日发表的119项研究(PubMed 31项、Embase 45项、Web of Science 43项),经预设纳入和排除标准后纳入20项分析。
- 纳入研究使用多种机器学习算法和影像组学方法从CT和MRI影像数据中提取定量特征。
诊断标准与关键数值
- 常用算法包括逻辑回归、极端梯度提升、支持向量机、随机森林和神经网络。
临床应用与阅片要点
- 影像组学特征与临床数据联合建模可提高骨折风险预测准确性,阅片时可关注椎体骨质量相关定量特征的提取与报告。
局限性
- 多数研究以女性为主且聚焦椎体,人群和部位覆盖有限。
- 未来需更大规模多中心数据集以提升模型稳健性和临床适用性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
AIM: Osteoporosis (OP) is a prevalent metabolic bone disease causing millions of fractures annually, leading to significant healthcare and economic burdens. Early diagnosis and intervention are critical, particularly among postmenopausal women and the elderly. Traditional bone mineral density tests, while useful, are limited in predicting fracture risk and are not always accessible. This review summarises recent advancements in applying machine learning (ML) and radiomics techniques for fracture risk prediction and OP assessment using computed tomography and magnetic resonance imaging.
MATERIALS AND METHODS: A comprehensive literature search was conducted in PubMed, Embase, and Web of Science using the keywords 'radiomics', 'osteoporosis', and 'fracture'. A total of 119 studies published between January 15, 2021, and March 15, 2025, were identified through PubMed (31), Embase (45), and Web of Science (43). After applying predefined inclusion and exclusion criteria, 20 studies were included for analysis. These studies used various ML algorithms and radiomics approaches to extract quantitative features from imaging data.
RESULTS: Most studies included predominantly female participants and focussed on vertebral bones. Integrating radiomics features with clinical data improved the predictive accuracy of fracture risk models. Commonly used algorithms included logistic regression, extreme gradient boosting, support vector machines, random forests, and neural networks. Fourteen studies achieved area under the curve (AUC) or coefficient of determination (R2) values ≥0.800.
CONCLUSION: ML and radiomics show great potential for enhancing OP and fracture risk assessment. Future studies should emphasise larger, multicentre datasets and broader clinical integration to improve model robustness and clinical applicability.
引用
- Jayasuriya AC, Skie M. Use of artificial intelligence and radiomics for diagnosis and prediction of osteoporotic fractures: a systematic review. Clin Radiol. 2026. PMID: 41955646. doi:10.1016/j.crad.2026.107308
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。