基于空间-治疗反应异质性分析的肿瘤疗效预测模型构建方法

The prediction of tumor treatment outcome can assist the individualized treatment planning, and improve the life and survival quality of patients. The existing prediction methods mainly take advantages of the tumor spatial heterogeneity information before treatment. Through the previous research conducted by the applicant, we found that the tumor treatment response heterogeneity is strongly correlated with the treatment outcome. We hypothesize the imaging features that represent the tumor treatment response heterogeneity could be important supplementary information for the prediction of tumor treatment outcome, and the prediction model which combined tumor spatial heterogeneity and treatment response heterogeneity would have better predictive ability. To validate the proposed hypothesis, we will build a prediction model based on tumor spatial and treatment response heterogeneity for the prediction of castration-resistant prostate cancer (CRPC). On the basis of multiparametric magnetic resonance imaging (MP-MRI), we will use image registration, radiomics and deep learning techniques to extract imaging features of tumor spatial and treatment response heterogeneity. Through the comparison of differences between CRPC prediction models which based on single heterogeneity features and combined two heterogeneity features, the influence of treatment response heterogeneity in building prediction model will be investigated, and the results of the new model building method will be validated. This project will provide a novel model building strategies for tumor outcome prediction and improve the predictive accuracy, which may help in individualizing treatment.

预测肿瘤疗效，可辅助临床制定个体化治疗方案，提升患者生存质量与生存期。现有预测方法主要借助治疗前肿瘤影像的空间异质性信息。申请人前期研究发现，肿瘤治疗反应异质性与疗效有较强相关性。我们推测，表达肿瘤治疗反应异质性的影像特征可作为疗效预测的重要补充信息，结合肿瘤空间与治疗反应异质性进行建模可提高预测效果。为验证上述假说，本课题以前列腺癌为例，针对去势抵抗（CRPC）发生时间预测，建立一种基于空间-治疗反应异质性的疗效预测模型。借助多参数磁共振成像（MP-MRI），采用图像配准、影像组学和深度学习，从治疗前和治疗早期影像中分别提取肿瘤空间异质性和治疗反应异质性特征，通过比较单一异质性和两种异质性融合构建的CRPC发生时间预测模型的差异，明确治疗反应异质性对预测模型准确性的影响，验证模型构建新方法的效果。本研究有望为肿瘤疗效预测提供全新建模思路，提升预测准确性，为个性化精准医疗提供理论依据。

构建肿瘤诊断和疗效预测模型，可辅助临床制定个体化治疗方案，提升患者生存质量与生存期。本课题以前列腺癌的诊断和疗效预测为例，收集整理了与前列腺癌去势抵抗（CRPC）发生紧密相关的肿瘤良恶性指标及CRPC的独立预测因子——手术治疗后的包膜突破和切缘阳性等疗效指标。针对上述指标的术前诊断和预测难题，基于多参数磁共振（MP-MRI）影像数据，采用智能影像计算方法，定量分析肿瘤异质性特征，对各类特征提取与分类算法进行了比较实验，明确了针对肿瘤诊断或疗效预测任务最佳的特征筛选方法及建模方法，提出了基于决策融合和瘤周影像的肿瘤诊断及疗效预测建模策略，构建了前列腺肿瘤诊断及疗效预测模型，并成功将本项目开发的建模策略拓展应用于其他肿瘤的诊断和疗效预测中，构建了脑肿瘤、卵巢癌、肺癌、直肠癌、肝癌等诊断和疗效预测模型，充分验证了所提建模策略的临床应用价值。

内容获取失败，请点击重试