评估脑动脉瘤破裂风险的预测模型
该研究对129个颅内动脉瘤行高分辨率磁共振成像前瞻性扫描,构建结合临床信息、形态学指标、瘤壁强化及影像组学的四种预测模型,比较其预测症状性动脉瘤的表现。最佳模型纳入年龄和影像组学数据,AUC达0.87。
要点
- 共纳入129个颅内动脉瘤,其中34个(26%)在就诊时为破裂或症状性。
- 基于PHASES评分的临床模型AUC为0.62(准确率70%,敏感度44%,特异度79%)。
- 结合PHASES评分与大小比的模型AUC提升至0.79(准确率73%,敏感度77%,特异度72%)。
核心结论
- 动脉瘤的分诊过程高度个体化,可受益于整合临床数据、动脉瘤详细形态学指标及瘤壁强化的复杂分析如影像组学。
研究设计与样本
- 前瞻性扫描颅内动脉瘤患者行高分辨率磁共振成像,分析磁共振图像并生成三维强化图以分析瘤壁强化,同时从瘤壁提取影像组学特征。
- 创建四种结合临床患者信息、动脉瘤形态学指标、瘤壁强化和影像组学数据的预测模型,比较其预测症状性动脉瘤表现的表现。
- 共纳入129个颅内动脉瘤,其中34个(26%)在就诊时为破裂或症状性。
主要发现
- 利用临床变量基于PHASES评分的临床模型检测症状性动脉瘤的AUC为0.62(准确率70%,敏感度44%,特异度79%)。
- 结合PHASES评分与形态学动脉瘤信息(大小比)的联合模型将AUC提升至0.79(准确率73%,敏感度77%,特异度72%)。
- 加入动脉瘤瘤壁强化指标进一步将模型AUC提升至0.82(准确率81%,敏感度65%,特异度86%)。
- 表现最佳的模型AUC达0.87(准确率76%,敏感度88%,特异度72%),纳入年龄和影像组学数据。
局限性
- 原文未报告局限性
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)。
原文
OBJECTIVE: Increased contrast enhancement of the aneurysm wall may indicate aneurysm instability. The authors tested different predictive models of aneurysm instability on a cohort of patients imaged with high-resolution magnetic resonance imaging (HR-MRI). METHODS: Patients with intracranial aneurysms were prospectively scanned with HR-MRI. Aneurysm instability was defined as rupture or symptomatic status at presentation. Magnetic resonance images were analyzed, and 3D enhancement maps were generated to analyze aneurysm wall enhancement. Additionally, radiomics features were extracted from the aneurysm wall. Four different predictive models combining clinical patient information, morphological aneurysm metrics, wall enhancement, and radiomics data were created to compare their performance in predicting symptomatic aneurysm presentation. RESULTS: A total of 129 intracranial aneurysms were included, with 34 (26%) being ruptured or symptomatic at presentation. The clinical model utilizing clinical variables and based on the Population, Hypertension, Age, Size of aneurysm, Earlier subarachnoid hemorrhage of another aneurysm, and Site of aneurysm (PHASES) score achieved an area under the curve (AUC) of 0.62 (70% accuracy, 44% sensitivity, and 79% specificity) in detecting symptomatic aneurysms. The combined model with the PHASES score and morphological aneurysm information (size ratio) improved the AUC to 0.79 (73% accuracy, 77% sensitivity, and 72% specificity). Adding aneurysm wall enhancement metrics further enhanced the model's performance, raising the AUC to 0.82 (81% accuracy, 65% sensitivity, and 86% specificity). The best performing model achieved an AUC to 0.87 (76% accuracy, 88% sensitivity, and 72% specificity) and included age and radiomic data. CONCLUSIONS: The triage process for aneurysms is highly personalized and can benefit from incorporating clinical data, detailed morphological metrics of the aneurysm, and sophisticated analyses of aneurysm wall enhancement such as radiomics.
引用
- Sagues E, Gudino A, Dier C, et al. Predictive models for assessing the risk of brain aneurysm rupture. J Neurosurg. 2025. PMID: 40315620. doi:10.3171/2025.1.JNS241772
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。