动态对比增强磁共振成像联合人工智能预测乳腺癌新辅助化疗疗效的价值分析
Value analysis of DCE-MRI-AI in predicting the curative effect of neoadjuvant chemotherapy for breast cancer
龚俊峰 1王永杰 1丁宇 1薛周 1沈高耀1
作者信息
- 1. 上海新华医院崇明分院医学影像科 上海 202150
- 折叠
摘要
目的:探讨基于人工智能(AI)技术动态对比增强磁共振成像(DCE-MRI)预测乳腺癌新辅助化疗疗效的价值.方法:收集2018年1月至2020年12月上海新华医院崇明分院收治的进行新辅助化疗治疗的89例乳腺癌患者,依据治疗效果将其分为有效组(36例)和无效组(53例),新辅助化疗前和第4疗程结束后进行DCE-MRI检查,绘制时间-信号强度曲线,测定容量转移常数(Ktrans)、速率常数(Kep)以及血管外间隙容积比(Ve).扩充原始DCE-MRI图像,提取包含病灶的感兴趣区域,运用深度卷积神经网络进行卷积运算,通过训练集获得分类模型.绘制受试者工作特征(ROC)曲线,选择ROC曲线下面积(AUC)最高的验证集模型作为最终模型,以测试集评价模型性能.结果:化疗后,DCE-MRI检查,有效组患者Ktrans(0.67±0.15)、Kep(1.22±0.24)显著小于无效组,ΔKtrans(1.51±0.18)、ΔKep(2.31±0.26)显著大于无效组,差异均有统计学意义(t=23.072、20.016;P＜0.05);ΔKtrans、ΔKep及ΔVe预测乳腺癌新辅助化疗疗效的AUC分别为0.814、0.839和0.432,三者联合检测的灵敏度、特异度及AUC分别为89.93%、83.48%和0.845,预测价值高于单项检测.DCE-MRI-AI模型对乳腺癌新辅助化疗疗效具有较高的预测效能,其训练集、验证集及测试集的AUC分别为0.897、0.869和0.859.在训练集和验证集中,DCE-MRI-AI模型和DCE-MRI模型对乳腺癌新辅助化疗疗效的预测效能差异具有统计学意义(Z训练集=2.435,P＜0.05;Z验证集=2.147,P＜0.05).结论:AI技术应用于DCE-MRI有助于提高对乳腺癌新辅助化疗疗效的预测效能,为治疗及预后评估提供可靠数据,具有临床应用价值.
Abstract
Objective:To investigate the value of dynamic contrast enhanced magnetic resonance imaging(DCE-MRI)based on artificial intelligence(AI)technique in predicting the curative effect of neoadjuvant chemotherapy for breast cancer.Methods:A total of 89 patients with breast cancer received neoadjuvant chemotherapy in Chongming Branch of Shanghai Xinhua Hospital from January 2018 to December 2020 were collected,and they were divided into an effectiveness group(36 cases)and an ineffectiveness group(53 cases)according to the treatment effect.DCE-MRI examinations were performed before neoadjuvant chemotherapy and after the end of the fourth course of treatment,and a time-signal intensity curves were drawn.Volume transfer constant(Ktrans),rate parameter(Kep)and extravascular space volume ratio(Ve)were measured.The original DCE-MRI image was expanded,and the region of interest containing the lesion was extracted,and the deep convolutional neural network was used to carry out convolutional operation,and the classification model was obtained through training set.The receiver operating characteristic(ROC)curve was drawn,and the model with the highest value of area under curve(AUC)of validation set was selected as the terminal model to evaluate the model performance of the test set.Results:After chemotherapy,the results of DCE-MRI showed that Ktrans(0.67±0.15)and Kep(1.22±0.24)in the effectiveness group were significantly lower than those in the ineffectiveness group,while ΔKtrans(1.51±0.18)and ΔKep(2.31±0.26)were significantly higher than those in the ineffectiveness group,and the differences were statistically significant(t=23.072,20.016,P<0.05).The AUC values of ΔKtrans,ΔKep and ΔVe in predicting the curative effect of neoadjuvant chemotherapy for breast cancer were respectively 0.814,0.839 and 0.432.The sensitivity,specificity and AUC value of the combined detection of them were respectively 89.93%,83.48%and 0.845,which predictive value was higher than that of single detection.The DCE-MRI-AI model has higher predictive efficiency for the curative effect of neoadjuvant chemotherapy for breast cancer,and the AUC values of training set,verification set and test set of that were respectively 0.897,0.869 and 0.859.In the training set and verification set,there was statistical difference between DCE-MRI-AI model and DCE-MRI model in predicting efficiency for the curative effect of neoadjuvant chemotherapy for breast cancer(Ztraining set=2.435,Zverification set=2.147,P<0.05).Conclusion:The application of AI technique in DCE-MRI is helpful to improve the predictive efficiency for neoadjuvant chemotherapy for breast cancer,and provide reliable data for the treatment and prognostic assessment,which has clinical application value.
关键词
乳腺癌/新辅助化疗/动态对比增强磁共振成像(DCE-MRI)/人工智能(AI)/卷积神经网络/疗效评价Key words
Breast cancer/Neoadjuvant chemotherapy/Dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)/Artificial intelligence(AI)/Convolutional neural network:Evaluation of curative effect引用本文复制引用
基金项目
上海市崇明区可持续发展科技创新行动计划(CKY2021-10)
出版年
2024
国家科技期刊平台
NSTL
万方数据