基于纵向时间影像动态增强模式分析的乳腺癌新辅助化疗疗效预测  

Prediction of Breast Cancer Neoadjuvant Chemotherapy Based on Dynamic Enhanced Mode Analysis of Longitudinal Time Images

作　　者：苏天放 范明 厉力华[1] Su Tianfang;Fan Ming;Li Lihua(Institute of Biomedical Engineering and Instrument,Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学生物医学工程与仪器研究所,杭州310018

出　　处：《中国生物医学工程学报》2023年第2期139-147,共9页Chinese Journal of Biomedical Engineering

基　　金：浙江省自然科学基金(LR23F010002);国家自然科学基金(62271178,U21A20521)。

摘　　要：新辅助化疗由于其较长的治疗周期,对化疗最终疗效早期准确的预测具有重要的临床参考价值。传统影像组学方法由于肿瘤异质性及影像部分容积效应等因素的存在,使得预测的精度难以进一步提高。本研究通过深度影像分解生成不同动态增强模式影像,对深度模式进行分析并基于纵向时间影像特征对新辅助化疗疗效进行预测分析。实验将采集的191例乳腺癌患者影像进行预处理,得到肿瘤和腺体感兴趣区域影像并分别提取影像组学特征,计算纵向时间特征变化率。使用随机森林模型对疗效进行预测分析并结合AUC指标对模型分类性能进行评估分析。结果表明,在分解前原始影像的预测任务中取得0.791的最佳AUC。在影像深度分解实验中,肿瘤影像的纵向模式变化在疗效组别中分布更具显著差异(P<0.01),在不同动态模式影像特征的预测任务中取得0.888的最佳AUC。综上,通过结合多区域影像和纵向时间特征,相比于分解前影像,深度分解后的不同模式影像进一步提升了基于特征水平的疗效预测能力,有望对患者早期诊断和方案调整提供重要的参考依据。Due to its long treatment cycle,neoadjuvant chemotherapy has important clinical reference value for early and accurate prediction of the final curative effect of chemotherapy.Due to the existence of factors such as tumor heterogeneity and image partial volume effect,the traditional radiomics method makes it difficult to further improve the prediction accuracy.Features were used to predict the efficacy of neoadjuvant chemotherapy.In the experiment,images of 191 breast cancer patients collected were preprocessed to obtain images of regions of interest in tumors and glands,and radiomics features were extracted,and the longitudinal time feature change rate was calculated.The random forest model was used to predict the curative effect and combined with the AUC index to evaluate and analyze the classification performance of the model.The results showed that the best AUC of 0.791 was achieved in the task of predicting raw images before decomposition.In the image depth decomposition experiment,the distribution of longitudinal pattern changes in tumor images was more significantly different among treatment groups(P<0.01),and the best AUC of 0.888 was achieved in the prediction task of image features of different dynamic patterns.In summary,by combining the multi-regional images and longitudinal time features,compared with the images before decomposition,the images of different modes after deep decomposition improved the curative effect prediction ability based on the feature level,which was expected to provide important reference for early diagnosis of patients and program adjustment in accordance with.

关 键 词：乳腺癌 动态对比增强磁共振影像分解 新辅助化疗 纵向时间分析 

分 类 号：R318[医药卫生—生物医学工程]