摘要
目的基于全视野数字病理切片(WSIs)使用机器学习技术开发构建预测模型,快速、准确预测胶质瘤的IDH状态和1p19q共缺失状态,减少对传统分子检测的依赖。方法本研究为回顾性研究,纳入2011—2022年郑州大学第一附属医院2072例成人型弥漫性胶质瘤患者的WSIs及临床信息。使用CellProfiler软件从WSIs中提取形态学、纹理、颜色等多维度病理组学特征,并通过Z-score标准化处理。采用Boruta算法结合随机森林模型筛选关键特征集,分别用于IDH状态和1p19q共缺失状态预测。随后,使用随机森林算法构建预测模型,并通过10折交叉验证进行训练和优化。模型性能通过ROC曲线、PR曲线和校准曲线评估。此外,通过Kaplan-Meier曲线对比评估模型预测效能。结果IDH状态预测模型在训练集和验证集上的AUC分别为0.86和0.82,PR曲线训练集AUC为0.78,校准曲线显示预测概率与实际概率高度一致。1p19q共缺失预测模型在训练集和验证集上的AUC分别为0.82和0.77,PR曲线训练集AUC为0.52,校准曲线显示出较高的预测准确性。Kaplan-Meier生存分析显示,模型预测的KM曲线与真实曲线贴合紧密,验证了模型预测效能。结果表明,病理组学模型可成功预测胶质瘤的IDH状态和1p19q共缺失状态。结论成功构建基于WSIs的病理组学预测模型,可快速、准确预测胶质瘤IDH状态和1p19q共缺失状态,具有临床应用潜力。
Objective To develop a predictive model based on whole-slide images(WSIs)using machine learning technology for the rapid and accurate prediction of IDH status and 1p19q co-deletion status in gliomas,thereby reducing reliance on traditional molecular testing.Methods This retrospective study included WSIs and clinical information from 2072 adult patients with diffuse gliomas treated at the First Affiliated Hospital of Zhengzhou University between 2011 and 2022.Morphological,textural,and color-based pathomics features were extracted from WSIs using CellProfiler software and normalized using Z-score standardization.The Boruta algorithm combined with a random forest model was employed to select key feature sets for predicting IDH status and 1p19q co-deletion status.Subsequently,random forest algorithms were used to construct predictive models,which were trained and optimized through 10-fold cross-validation.Model performance was evaluated using ROC curves,PR curves,and calibration curves.Additionally,Kaplan-Meier curves were used to compare and assess the predictive efficacy of the models.Results The IDH status prediction model achieved AUCs of 0.86 and 0.82 in the training and validation sets,respectively,with a PR curve AUC of 0.78 in the training set.The calibration curve demonstrated high consistency between predicted and actual probabilities.The 1p19q co-deletion prediction model achieved AUCs of 0.82 and 0.77 in the training and validation sets,respectively,with a PR curve AUC of 0.52 in the training set.The calibration curve indicated high predictive accuracy.Kaplan-Meier survival analysis revealed that the model-predicted KM curves closely aligned with the true curves,validating the model′s predictive efficacy.The results demonstrate that the pathomics model can successfully predict IDH status and 1p19q co-deletion status in gliomas.Conclusion This study successfully developed a WSI-based pathomics predictive model capable of rapidly and accurately predicting IDH status and 1p19q co-deletion status in gliomas,demonstrating potential for clinical application.
作者
任悠悠
王伟伟
REN You-you;WANG Wei-wei(Basic Medical College,Zhengzhou University,Zhengzhou 450052,China;Department of Pathology,The First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China)
出处
《南阳理工学院学报》
2025年第4期109-116,共8页
Journal of Nanyang Institute of Technology
