目的 探讨多参数MRI联合影像组学在区分前列腺影像报告数据系统(PI-RADS)4～5分病灶良恶性中的应用价值,旨在构建能够有效区分两者的模型,提高诊断精确度.材料与方法 回顾性分析2018年1月至2021年6月在我院行前列腺多参数MRI(multiparametric MRI,mpMRI)检查后PI-RADS评分为4～5分的患者病例共135例,其中病理诊断良性64例,恶性71例.将入组病例随机分层采样按照7∶3的比例划分为训练集(95例,良性45例,恶性50例)和测试集(40例,良性19例,恶性 21 例),分析前列腺特异性抗原(prostate specific antigen,PSA)、MRI传统参数表观弥散系数(apparent diffusion coefficient,ADC)、多模态影像组学模型[T2WI、弥散加权成像(diffusion weighted imaging,DWI)、ADC三种模态]及联合模型(多模态影像组学联合传统参数ADC值)对PI-RADS 4～5分病灶良恶性的鉴别效力,构建诊断模型.结果 良性病变组ADC值[(0.791±0.149)×10-3 mm2/s]显著高于恶性组[(0.612±0.110)×10-3 mm2/s],其在训练集、测试集中的曲线下面积(area under the curve,AUC)分别为0.870、0.772.而良性病变组总PSA(total PSA,tPSA)略低于恶性组,两组之间差异无统计学意义.ADC、DWI和T2WI多模态组合影像组学在训练集、测试集中的AUC分别为0.942、0.850.联合模型在训练集、测试集中的AUC分别为0.952、0.842.结论 综合多模态MRI(T2WI、DWI、ADC)影像组学联合传统参数的模型能有效帮助鉴别PI-RADS 4～5分病灶的良恶性,帮助提高诊断准确度,进一步辅助个体化治疗方案的制订.
Predicting malignancy of PI-RADS 4-5 lesions with radiomics features based on multiparametric magnetic resonance imaging
Objective:This study was aimed to distinguish benign and malignant prostate lesions prostate imaging reporting and data system(PI-RADS)4-5 scored using multiparametric MRI(mpMRI)combined with imaging radiomics,and to construct a diagnostic model for predicting malignancy of prostate lesions scored PI-RADS 4-5.Materials and Methods:Clinical,pathological and imaging data of patients who underwent prostate mpMRI examination and scored PI-RADS 4-5 in our hospital from January 2018 to June 2021 were retrospectively collected.A total of 135 patients were enrolled,including 64 benign and 71 malignant cases.They were then randomly divided into training set(n=95,45 benign cases,50 malignant cases)and test set(n=40,19 benign cases,21 malignant cases).Different radiomics models were used for identification.Results:Higher apparent diffusion coefficient(ADC)value and lower total prostate specific antigen(tPSA)concentration were observed in benign group.ADC values in benign and malignant groups were(0.791±0.149)and(0.612±0.110)×10-3 mm2/s,respectively,with cut-off value of 0.712×10-3 mm2/s,area under the curve(AUC)values in training set and test set were 0.870 and 0.772,respectively.The AUC of radiomics model based on mpMRI[ADC,diffusion weighted imaging(DWI)and T2WI]in training set and test set was 0.942 and 0.850,respectively.The AUC of multimodal radiomics model combined ADC value and radiomics score(Radscore)in the training set and test set was 0.952 and 0.842,respectively.Conclusions:Traditional mpMRI parameters combined with imaging features did great job in identifying benign and malignant prostate PI-RADS 4-5 lesions,which shows immense potential in further clinical practice by assisting the formulation of subsequent individual treatment strategies.
prostate cancerradiomicsmagnetic resonance imagingprostate imaging reporting and data system
国家科技期刊平台
NSTL
万方数据
维普
