脑积水影像学进展:从二维指标到三维生物标志物
本文为叙述性综述,经PubMed检索脑积水影像生物标志物、脑室体积测量、自动分割、影像组学与机器学习研究,归纳二维线性指标的结构局限,并综述三维体积与影像组学在小儿脑积水监测、脑室扩大鉴别诊断、脑脊液分流术前反应预测及术后纵向随访四个领域的应用。
要点
- 三维体积与影像组学方法在上述四个临床领域的表现优于线性指标
- 机器学习模型在鉴别诊断与分流反应预测中报告AUC超过0.9
- 临床转化受限于单中心小样本、缺乏外部验证及未定义体积变化阈值
核心结论
- 三维体积与影像组学方法在四个脑积水临床领域中一致优于线性指标,但临床转化仍受证据与阈值限制。
研究设计与样本
- 叙述性文献综述,通过PubMed检索脑积水影像生物标志物、脑室体积测量、自动分割、影像组学与机器学习应用研究,并手工检索参考文献列表补充来源。
主要发现
- 在四个临床领域中,三维体积与影像组学方法一致优于线性指标。
- 机器学习模型在鉴别诊断与分流反应预测中报告AUC超过0.9。
- 自动分割在检测任务中已达到优异性能,体积测量对术后变化比Evans指数更敏感。
关键数值与推荐
- Evans指数、胼胝体角、额枕角比等二维线性指标仍是脑积水评估的临床标准,但受测量变异、对脑脊液空间重分布不敏感及对体积变化敏感性下降的限制。
- 机器学习模型在鉴别诊断与分流反应预测中报告AUC超过0.9。
- 目前尚未定义具有临床意义的体积变化阈值。
临床怎么用
- 三维体积与影像组学可作为二维线性指标的补充,用于小儿脑积水监测、脑室扩大鉴别、分流术前预测与术后随访。
- 临床常规应用前需将经过验证的工具嵌入影像工作流程与临床指南。
局限性
- 多为单中心小样本数据集。
- 缺乏外部验证。
- 临床有意义的体积变化阈值尚未定义。
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)。
原文
PURPOSE: Linear two-dimensional indices such as the Evans index, callosal angle, and fronto-occipital horn ratio remain the clinical standard for hydrocephalus assessment, yet are limited by measurement variability, insensitivity to spatial CSF redistribution, and reduced sensitivity to volumetric change over time. This narrative review aims to summarize established two-dimensional indices, their structural limitations, and describe how automated segmentation and radiomic feature extraction enable three-dimensional assessment across four clinical domains. METHODS: A narrative literature review was conducted using PubMed. Studies addressing hydrocephalus imaging biomarkers, ventricular volumetry, automated segmentation, radiomics, and machine learning applications were reviewed. Reference lists of relevant articles were hand-searched for additional sources. The literature was synthesized across four clinical domains: pediatric hydrocephalus monitoring, differential diagnosis of ventriculomegaly, preoperative prediction of response to cerebrospinal fluid diversion, and longitudinal post-treatment follow-up. RESULTS: Across pediatric hydrocephalus monitoring, differential diagnosis of ventriculomegaly, preoperative prediction of response to CSF diversion, and post-shunt longitudinal follow-up, three-dimensional volumetric and radiomic approaches consistently outperform linear indices. Machine learning models report AUCs exceeding 0.9 for differential diagnosis and shunt response prediction. Automated segmentation has reached excellent performance for detection tasks, and volumetry is more sensitive to postoperative change than the Evans index. CONCLUSIONS: Despite strong metric performance, clinical translation remains limited by small single-centre datasets, missing external validation, and undefined thresholds for clinically meaningful volumetric change. Embedding validated tools into radiological workflows and clinical guidelines will be essential before three-dimensional biomarkers can improve routine hydrocephalus care.
引用
- Da Mutten R, Turczynski Holmgren R, Edström E, et al. Moving on in hydrocephalus imaging: from 2D to 3D biomarkers. Acta Neurochir (Wien). 2026. PMID: 42479243. doi:10.1007/s00701-026-06985-2
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。