摘要
目的:构建结合SHAP(Shapley Additive Explanations)可解释性分析和MRI影像组学的机器学习模型,为乳腺癌患者新辅助化疗的疗效提供准确和可解释的预测。方法:本研究入组60例乳腺癌患者,对ER、PR、HER-2和Ki-67状态进行了分析。进一步选取关键的组学特征,使用XGBoost构建了临床和组学集成模型,并通过SHAP方法对预测模型进行可解释性分析。结果:pCR和Non-pCR患者在ER、PR、HER-2状态上有显著差异,但Ki-67无显著差异。集成模型的AUC为0.972,准确率达到90.0%。SHAP分析显示,两个组学特征的重要性显著高于HER-2状态。结论:该研究成功结合SHAP与MRI影像组学构建了一个高准确性和具有可解释性的新辅助化疗预测模型,为临床评估乳腺癌患者新辅助化疗疗效提供了重要参考。
Objective:To explore the integration of SHAP interpretability analysis with MRI radiomics machine learning models to provide accurate and interpretable predictions for the efficacy of neoadjuvant chemotherapy in breast cancer patients.Methods:This study enrolled 60 breast cancer patients and analyzed their ER,PR,HER-2,and Ki-67 status.Key radiomic features were further selected,and an integrated clinical and radiomics model was constructed using XGBoost.The interpretability of the model was evaluated using SHAP analysis.Results:Significant differences were observed in ER,PR,and HER-2 status between pCR and Non-pCR patients,while no significant difference was noted for Ki-67.The integrated model achieved an AUC of 0.972 and an accuracy of 90.0%.SHAP analysis revealed that the importance of two radiomic features significantly surpassed that of HER-2 status.Conclusion:This study successfully established a highly accurate and interpretable predictive model for neoadjuvant chemotherapy by integrating SHAP with MRI radiomics,offering valuable insights for the clinical assessment of neoadjuvant chemotherapy in breast cancer patients.
作者
赵玲
陈正国
罗锐
尹龙洲
周莉
黄瑶
ZHAO Ling;CHEN Zhengguo;LUO Rui;YIN Longzhou;ZHOU Li;HUANG Yao(Mianyang Hospital Affiliated to Medical School of University of Electronic Science and Technology,Mianyang Central Hospital,Sichuan Mianyang 621000,China;School of Medicine,Chongqing University,Chongqing 400030,China.)
出处
《现代肿瘤医学》
CAS
2024年第15期2839-2844,共6页
Journal of Modern Oncology
基金
2023年重庆市研究生科研创新项目(编号:CYB23070)。
关键词
乳腺癌
疗效预测
新辅助化疗
机器学习
breast cancer
therapeutic effect prediction
neoadjuvant chemotherapy
machine learning
