前列腺癌根治术后生化复发相关MRI影像组学特征的研究

MRI-based Radiomics Features Associated With Biochemical Recurrence of Prostate Cancer After Radical Prostatectomy

目的基于双参数磁共振成像(biparametric magnetic resonance imaging,bp-MRI)探讨影响前列腺癌根治术(radical prostatectomy,RP)后生化复发(biochemical recurrence,BCR)的影像组学特征。方法回顾性分析我科2010年1月~2016年6月278例RP患者的临床资料,术后至少随访3年,86例(30.9%)发生BCR。以患者术后病理切片为参考,采用认知融合的方式在术前mpMRI相应层面的T2WI和表观弥散系数(apparent diffusion coefficient,ADC)图片上勾画前列腺癌灶轮廓,获得感兴趣区(region of interest,ROI),对MRI片ROI内的影像组学特征进行提取。最小绝对选择与收缩算子(least absolute shrinkage and selection operator,LASSO)回归联合Cox回归模型用于影像组学特征降维、筛选及影像组学标签构建。单因素和多因素Cox回归分析影响BCR的预后因素。结果由T2WI和ADC图中各获取444个影像组学特征,共888个影像组学特征。通过LASSO回归进行特征降维及筛选,最终获得6个非零参数的影像组学特征。基于上述特征,结合Cox回归模型构建影像组学标签。多因素Cox回归分析显示影像组学标签(HR=2.404,95%CI:1.543~3.747,P=0.000)与国际泌尿病理协会(International Society of Urological Pathology,ISUP)分级(HR=1.235,95%CI:1.027~1.486,P=0.025)是BCR的独立预后因素。结论共获得6个非零系数的bp-MRI影像组学特征,以此构建的影像组学标签与RP后BCR显著相关。该标签可能有助于辅助识别RP治疗后BCR高风险的前列腺癌患者。

Objective To evaluate the biparametric magnetic resonance imaging(bp-MRI)based radiomics features associated with the biochemical recurrence(BCR)after radical prostatectomy(RP).Methods A total of 278 patients with prostate cancer who underwent RP in our department from January 2010 to June 2016 with a minimum of 3-year follow-ups were retrospectively investigated.The preoperative bp-MRI images of T2WI and apparent diffusion coefficient(ADC)at the corresponding level were identified by cognitive fusion with the reference of a pathological section to outline the tumor,and subsequently the region of interest(ROI)was acquired.Radiomic features in the ROI were extracted.The least absolute shrinkage and selection operator(LASSO)regression algorithm combined with the Cox regression model was used for feature dimensionality reduction,selection,and signature construction.Univariate and multivariate Cox regression were used to identify the prognostic factors of BCR.Results A total of 888 radiomics features were acquired from T2WI and ADC images(444 features for each).Through the LASSO regression for feature dimensionality reduction and selection,6 radiomics features with non-zero coefficients were identified.Based on the above features,the Cox regression model was utilized to construct a radiomics signature.In the multivariate analysis,the radiomics signature(HR=2.404,95%CI:1.543-3.747,P=0.000)and International Society of Urological Pathology(ISUP)grade(HR=1.235,95%CI:1.027-1.486,P=0.025)were independent prognostic factors of BCR.Conclusions Six imaging features with non-zero coefficients and the radiomics signature built from them are significantly correlated with BCR after RP.This signature may help identify prostate cancer patients with high risk of BCR following RP.