深度学习驱动的多模态融合模型用于预测大脑中动脉动脉瘤破裂风险
本研究回顾性纳入内部队列578个及两个外部验证集共51个大脑中动脉动脉瘤,基于原始CTA图像、影像组学特征、临床参数和形态学特征构建多模态融合深度学习模型MCANet预测破裂风险;内部队列中MCANet准确率91.38%、敏感度96.33%、精确率90.52%、F1分数93.33%,外部验证在两个独立中心保持良好区分能力。
要点
- MCANet融合原始CTA图像、影像组学、临床及形态学特征预测大脑中动脉动脉瘤破裂风险。
- 内部队列中MCANet准确率91.38%、敏感度96.33%、精确率90.52%、F1分数93.33%。
- 动脉瘤破裂独立预测因素包括多发动脉瘤、动脉瘤位置、动脉瘤角度、子囊及高宽比。
核心结论
- MCANet模型有效整合多模态异质数据预测大脑中动脉动脉瘤破裂风险,在数据受限场景下也显示出临床适用性,并可能通过个体化风险评估优化治疗决策、减轻患者焦虑。
研究设计与样本
- 回顾性研究,内部队列578个大脑中动脉动脉瘤,两个外部验证数据集共51个动脉瘤;采用多变量逻辑回归分析确定破裂独立预测因素;基于重建CTA图像量化动脉瘤形态学参数,并通过计算分析提取影像组学特征;构建整合原始CTA图像、影像组学特征、临床参数和形态学特征的多模态分类模型MCANet;外部验证使用来自两个独立医学中心的数据集评估模型泛化性和小样本稳健性;采用准确率、F1分数、精确率和召回率四项指标评估模型性能。
诊断标准与关键数值
- 内部队列中369个动脉瘤发生破裂。
- 破裂独立预测因素包括:存在多发动脉瘤、动脉瘤位置、动脉瘤角度、存在子囊动脉瘤及高宽比。
- MCANet在内部队列的预测性能:准确率91.38%、敏感度96.33%、精确率90.52%、F1分数93.33%。
- 外部验证在两个独立队列中均保持良好区分能力。
临床应用与阅片要点
- MCANet模型可整合原始CTA图像、影像组学特征、临床参数和形态学特征用于大脑中动脉动脉瘤破裂风险分层。
- 模型在数据受限场景下显示出临床适用性,可能通过个体化风险评估优化治疗决策并减轻患者焦虑。
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)。
英文原文(摘要)· 点击展开
RATIONALE AND OBJECTIVES: The decision to treat unruptured intracranial aneurysms remains a clinical dilemma. Middle cerebral artery (MCA) aneurysms represent a prevalent subtype of intracranial aneurysms. This study aims to develop a multimodal fusion deep learning model for stratifying rupture risk in MCA aneurysms.
MATERIALS AND METHODS: We retrospectively enrolled internal cohort and two external validation datasets with 578 and 51 MCA aneurysms, respectively. Multivariate logistic regression analysis was performed to identify independent predictors of rupture. Aneurysm morphological parameters were quantified using reconstructed CT angiography (CTA) images. Radiomics features of aneurysms were extracted through computational analysis. We developed MCANet - a multimodal data-driven classification model integrating raw CTA images, radiomics features, clinical parameters, and morphological characteristics - to establish an aneurysm rupture risk assessment framework. External validation was conducted using datasets from two independent medical centers to evaluate model generalizability and small-sample robustness. Four key metrics, including accuracy, F1-score, precision, and recall, were employed to assess model performance.
RESULTS: In the internal cohort, 369 aneurysms were ruptured. Independent predictors of rupture included: the presence of multiple aneurysms, aneurysm location, aneurysm angle, presence of daughter-sac aneurysm, and height-width ratio. MCANet demonstrated satisfactory predictive performance with 91.38% accuracy, 96.33% sensitivity, 90.52% precision, and 93.33% F1-score. External validation maintained good discriminative ability across both independent cohorts.
CONCLUSION: The MCANet model effectively integrates multimodal heterogeneous data for MCA aneurysm rupture risk prediction, demonstrating clinical applicability even in data-constrained scenarios. This model shows potential to optimize therapeutic decision-making and mitigate patient anxiety through individualized risk assessment.
引用
- Jia X, Chen Y, Zheng K, et al. Deep Learning-Driven Multimodal Fusion Model for Prediction of Middle Cerebral Artery Aneurysm Rupture Risk. Acad Radiol. 2025. PMID: 40691114. doi:10.1016/j.acra.2025.07.001
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。