引用本文
  • 臧书芹,沈浮,郝强*.磁共振影像组学主成分分析法预测直肠癌新辅助放射化学治疗效果[J].第二军医大学学报,2020,41(3):325-329    [点击复制]
  • ZANG Shu-qin,SHEN Fu,HAO Qiang*.Magnetic resonance imaging-based radiomics of rectal cancer: principal component analysis to predict treatment response after neoadjuvant chemoradiotherapy[J].Acad J Sec Mil Med Univ,2020,41(3):325-329   [点击复制]
【打印本页】 【下载PDF全文】 【HTML】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 201次   下载 187 本文二维码信息
码上扫一扫!
磁共振影像组学主成分分析法预测直肠癌新辅助放射化学治疗效果
臧书芹,沈浮,郝强*
0
(海军军医大学(第二军医大学)长海医院影像医学科, 上海 200433
*通信作者)
摘要:
目的 探讨基于MRI高分辨率T2加权图像的影像组学主成分分析(PCA)法对直肠癌新辅助放射化学治疗效果的预测价值。方法 回顾性分析2018年1月1日至2018年12月31日在我院因局部进展期直肠癌接受新辅助放射化学治疗后行直肠癌根治性切除术的80例患者资料,男60例、女20例,年龄为28~74岁,平均年龄为(56.2±9.9)岁。患者行新辅助放射化学治疗前接受3.0 T MRI检查,在高分辨率T2加权图像上提取影像组学特征,再采用PCA法进行特征值降维,使用降维后的特征与病理完全缓解(pCR)标签建立logistic回归分类器模型,将样本随机分为训练集与测试集进行机器学习,分别绘制ROC曲线并计算AUC及灵敏度、特异度、准确度。结果 MRI高分辨率T2加权图像共提取到1 409个影像组学特征,PCA法重新组合并选取了前5个最能代表整个影像组学特征矩阵的新特征,分别能代表整个影像组学特征矩阵中9.926 016 67×10-1、4.854 545 00×10-3、2.509 013 91×10-3、2.489 032 30×10-5、7.372 984 50×10-6的信息。Logistic回归分类器模型交叉验证测试集的平均AUC为0.761(95% CI:0.694~0.828),灵敏度为90.3%,特异度为40.0%,准确度为79.0%。结论 基于MRI高分辨率T2加权影像组学PCA法对直肠癌新辅助放射化学治疗的疗效具有较好的预测价值。
关键词:  直肠肿瘤  新辅助放射化学治疗  影像组学  磁共振成像  主成分分析
DOI:10.16781/j.0258-879x.2020.03.0325
投稿时间:2019-08-12修订日期:2019-12-23
基金项目:海军军医大学(第二军医大学)青年启动基金(2018QN05).
Magnetic resonance imaging-based radiomics of rectal cancer: principal component analysis to predict treatment response after neoadjuvant chemoradiotherapy
ZANG Shu-qin,SHEN Fu,HAO Qiang*
(Department of Radiology, Changhai Hospital, Naval Medical University(Second Military Medical University), Shanghai 200433, China
*Corresponding author)
Abstract:
Objective To explore the value of radiomics with principal component analysis (PCA) based on magnetic resonance imaging (MRI) high-resolution T2-weighted images for predicting treatment response after neoadjuvant chemoradiotherapy (nCRT) of rectal cancer. Methods Clinical data of 80 patients with locally advanced rectal cancer, who received nCRT before radical resection from Jan. 1, 2018 to Dec. 31, 2018 in our hospital, were retrospectively analyzed. There were 60 males and 20 females with a mean age of (56.2±9.9) years (range, 28-74 years). All the patients underwent 3.0 T MRI examination before nCRT. The radiomics features were extracted from the high-resolution T2-weighted images. PCA was applied to reduce the dimension. The logistic regression classifier model was built using the dimension-reduced feature and pathologic complete response label. The samples were randomly divided into training set and testing set for machine learning. ROC curves were drawn and AUC, sensitivity, specificity and accuracy were calculated. Results A total of 1 409 radiomics features were extracted from MRI high-resolution T2-weighted images. Five optimal features, which could best represent the overall radiomics feature matrix, were recombined and selected by PCA method. They represented the information on 9.926 016 67×10-1, 4.854 545 00×10-3, 2.509 013 91×10-3, 2.489 032 30×10-5 and 7.372 984 50×10-6, respectively. The mean AUC of the logistic regression classifier model was 0.761 (95% CI 0.694-0.828), sensitivity was 90.3%, specificity was 40.0%, and accuracy was 79.0%. Conclusion Radiomics with PCA based on MRI high-resolution T2-weighted images has good predictive value of treatment response after nCRT of rectal cancer.
Key words:  rectal neoplasms  neoadjuvant chemoradiotherapy  radiomics  magnetic resonance imaging  principal component analysis