多模态病理组学与临床特征预测儿童髓母细胞瘤术后永久性脑积水
该研究纳入2015年1月至2024年9月在珠江医院治疗的90例18岁以下髓母细胞瘤患者,按7:3随机分为训练集和验证集。利用ResNet-18模型从H&E染色切片提取病理组学特征,构建临床、病理组学及多模态三种逻辑回归模型预测术后永久性脑积水。验证集上多模态模型AUC达0.849(95%CI: 0.622-1.000),平均精度0.816(95%CI: 0.535-1.000),并改善风险分层(预测高风险组3年无脑积水生存率37.7% vs 低风险组89.5%,P<0.001)。
要点
- 多模态模型在验证集预测术后永久性脑积水的AUC为0.849。
- 预测高风险组3年无脑积水生存率37.7%,低风险组89.5%。
- 深度学习病理组学特征可辅助术后脑积水风险分层。
核心结论
- 整合深度学习病理组学特征与临床变量可显著改善儿童髓母细胞瘤术后脑积水风险分层。
研究设计与样本
- 回顾性分析2015年1月至2024年9月珠江医院收治的90例18岁以下髓母细胞瘤患者,按7:3随机分为训练集和验证集。
- 使用ResNet-18模型从H&E染色切片提取定量病理组学特征,构建临床、病理组学及多模态三种逻辑回归预测模型。
- 采用ROC和精确召回曲线评估模型性能,Kaplan-Meier法和log-rank检验评估风险分层。
主要发现
- 90例患者中26例(28.8%)因术后永久性脑积水需行脑脊液分流。
- 验证集上多模态模型AUC为0.849(95%CI: 0.622-1.000),平均精度为0.816(95%CI: 0.535-1.000)。
- 多模态模型改善风险分层:预测高风险组3年无脑积水生存率37.7%,低风险组89.5%,P < 0.001。
临床怎么用
- 可基于多模态模型对儿童髓母细胞瘤术后患者进行脑积水风险分层,辅助个体化管理。
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: Predicting postoperative persistent hydrocephalus risk in pediatric medulloblastoma remains challenging using conventional clinical features. We investigated whether deep learning (DL) of pathomic features could improve postoperative hydrocephalus risk stratification. METHODS: We analyzed patients under 18 years old with medulloblastoma treated at Zhujiang Hospital from January 2015 to September 2024, randomly split 7:3 into training and validation sets. Using ResNet-18 model, we extracted quantitative histopathological features from H&E-stained slides. We developed three logistic regression models to predict postoperative persistent hydrocephalus: (1) clinical model, (2) pathomic model, and (3) multimodal model (clinical + pathomics features). Model performance was evaluated using ROC and precision-recall curves. Risk group stratification was assessed with Kaplan-Meier plots and log-rank tests. RESULTS: A total of 90 patients was included in the study, 26(28.8%) patients required CSF diversion due to permanent hydrocephalus following tumor resection. On the validation set, the multimodal model achieved the highest predictive performance with an AUC of 0.849 (95% CI: 0.622-1.000) and average precision of 0.816 (95% CI: 0.535-1.000). The multimodal model improved risk-group stratification (3-year hydrocephalus-free survival for predicted high-risk: 37.7% versus low-risk: 89.5%, P < 0.001). CONCLUSION: Integrating deep learning-derived pathomic features with clinical variables significantly improves postoperative hydrocephalus risk stratification in pediatric medulloblastoma patients, facilitating personalized management strategies and ultimately enhancing patient outcomes.
引用
- Zhong W, Li Z, Lv S, et al. Multimodal pathomics and clinical features predict postresection permanent hydrocephalus in pediatric medulloblastoma. J Neurooncol. 2025. PMID: 41171470. doi:10.1007/s11060-025-05263-y
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。