文献速递前列腺癌

多参数磁共振成像测量的放射组学特征预测前列腺癌侵袭性

Radiomics Features Measured with Multiparametric Magnetic Resonance Imaging Predict Prostate Cancer Aggressiveness.
PubMed 文献 · J Urol · 2019-08-08

回顾性分析64例前列腺癌患者,从T2加权、表观扩散系数和扩散峰度成像图中提取226个放射组学特征,与病理Gleason评分、40个基因表达特征及698个前列腺癌相关基因表达水平进行相关性分析,并构建机器学习模型预测Gleason评分≥8和Decipher评分≥0.6。14个放射组学特征与Gleason评分显著相关(最高r=0.39,p=0.001),31个纹理和直方图特征与19个基因特征显著相关,40个弥散加权成像特征与132个基因表达水平显著相关。机器学习模型预测Gleason评分≥8的AUC为0.72,预测Decipher评分≥0.6的AUC为0.84。

要点

  • 14个放射组学特征与Gleason评分显著相关,最高相关系数r=0.39(p=0.001)。
  • 31个纹理和直方图特征与19个基因特征显著相关,其中与PORTOS特征相关性最强(r=-0.481,p=0.002)。
  • 机器学习模型预测Gleason评分≥8的AUC为0.72,预测Decipher评分≥0.6的AUC为0.84。

核心结论

  • 磁共振成像放射组学特征在前列腺癌的组织病理学和基因组学水平上是有前景的侵袭性标志物。

研究设计与样本

  • 回顾性分析64例前列腺癌患者,平均年龄64岁(范围41至76岁),于2016年1月至2017年1月期间在根治性前列腺切除术前接受磁共振成像检查。
  • 从T2加权、表观扩散系数和扩散峰度成像图中提取226个磁共振成像放射组学特征,包括直方图和纹理特征以及病灶大小和PI-RADS评分。
  • 构建基于放射组学特征的交叉验证、套索正则化逻辑回归机器学习模型,用于预测Gleason评分≥8和Decipher评分≥0.6。

主要发现

  • 14个放射组学特征与Gleason评分显著相关(最高相关系数r=0.39,p=0.001)。
  • 31个纹理和直方图特征与19个基因特征显著相关,其中与PORTOS特征相关性最强(r=-0.481,p=0.002)。
  • 40个弥散加权成像特征与132个基因表达水平显著相关。
  • 机器学习模型预测Gleason评分≥8的表现尚可(AUC 0.72),预测Decipher评分≥0.6的表现优秀(AUC 0.84)。

局限性

  • 原文未报告局限性

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: We sought to 1) assess the association of radiomics features based on multiparametric magnetic resonance imaging with histopathological Gleason score, gene signatures and gene expression levels in prostate cancer and 2) build machine learning models based on radiomics features to predict adverse histopathological scores and the Decipher® genomics metastasis risk score. MATERIALS AND METHODS: We retrospectively analyzed the records of 64 patients with prostate cancer with a mean age of 64 years (range 41 to 76) who underwent magnetic resonance imaging between January 2016 and January 2017 before radical prostatectomy. A total of 226 magnetic resonance imaging radiomics features, including histogram and texture features in addition to lesion size and the PI-RADS™ (Prostate Imaging Reporting and Data System) score, were extracted from T2-weighted, apparent diffusion coefficient and diffusion kurtosis imaging maps. Radiomics features were correlated with the pathological Gleason score, 40 gene expression signatures, including Decipher, and 698 prostate cancer related gene expression levels. Cross-validated, lasso regularized, logistic regression machine learning models based on radiomics features were built and evaluated for the prediction of Gleason score 8 or greater and Decipher score 0.6 or greater. RESULTS: A total of 14 radiomics features significantly correlated with the Gleason score (highest correlation r = 0.39, p = 0.001). A total of 31 texture and histogram features significantly correlated with 19 gene signatures, particularly with the PORTOS (Post-Operative Radiation Therapy Outcomes Score) signature (strongest correlation r = -0.481, p = 0.002). A total of 40 diffusion-weighted imaging features correlated significantly with 132 gene expression levels. Machine learning prediction models showed fair performance to predict a Gleason score of 8 or greater (AUC 0.72) and excellent performance to predict a Decipher score of 0.6 or greater (AUC 0.84). CONCLUSIONS: Magnetic resonance imaging radiomics features are promising markers of prostate cancer aggressiveness on the histopathological and genomics levels.

来源:J Urol,查看原文。版权归原作者所有。

引用

  1. Hectors SJ, Cherny M, Yadav KK, et al. Radiomics Features Measured with Multiparametric Magnetic Resonance Imaging Predict Prostate Cancer Aggressiveness. J Urol. 2019. PMID: 30958743. doi:10.1097/JU.0000000000000272

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。