目的探讨基于T2WI影像组学列线图鉴别诊断卵巢成人型颗粒细胞瘤与扩散加权成像(diffusion weighted imaging,DWI)高信号纤维-卵泡膜细胞肿瘤的效能.材料与方法回顾性收集北京大学第三医院2019年1月至2023年10月经手术病理确诊的卵巢成人型颗粒细胞瘤29例和DWI呈高信号的纤维-卵泡膜细胞肿瘤61例.所有肿瘤按7:3的比例随机分为训练集和验证集.应用单因素分析和多因素logistic回归筛选出临床和常规MRI征象,建立临床模型.基于T2WI提取影像组学特征,应用K最佳和最小绝对收缩和选择算法进行特征筛选,构建影像组学模型并计算影像组学评分(radiomics score,Rad-score).联合临床模型和Rad-score构建列线图模型.应用受试者工作特征曲线(receiver operating characteristic,ROC)分析各模型诊断效能,应用决策曲线分析(decision curve analysis,DCA)评价模型的临床价值.结果经logistic回归分析,将"蜂窝样"小囊变[比值比(odds ratio,OR)值=0.20,95％置信区间(confidence interval,CI)=0.05～0.79,P=0.022]和肿瘤内出血(OR值=0.16,95％CI=0.03～0.98,P=0.048)用于构建临床模型.基于T2WI筛选保留了9个组学特征用于构建影像组学模型.由"蜂窝样"小囊变、肿瘤内出血和Rad-score构建列线图模型.影像组学模型和列线图模型的ROC曲线下面积(area under the curve,AUC)均高于临床模型(训练集:0.983 vs.0.742,Z=−4.058,P<0.001;0.969 vs.0.742,Z=−3.817,P<0.001.验证集:0.858 vs.0.731,Z=−1.388,P=0.165;0.883 vs.0.731,Z=−1.612,P=0.107),列线图和影像组学模型的AUC差异无统计学意义(训练集:Z=−1.040,P=0.298;验证集:Z=0.822,P=0.411).DCA显示列线图和影像组学模型明显优于临床模型.结论本研究所构建的基于T2WI的影像组学模型和列线图模型能有效鉴别卵巢颗粒细胞瘤和DWI高信号纤维-卵泡膜细胞肿瘤,且效能优于基于常规MRI征象的临床模型.
T2WI-based radiomics for discriminating between ovarian adult-type granulosa cell tumor and ovarian fibroma-thecoma with high-signal intensity on DWI
Objective:To investigate the value of T2WI-based radiomics nomogram for the preoperative differentiation of ovarian adult-type granulosa cell tumor and ovarian fibroma-thecoma with high-signal intensity on diffusion weighted imaging (DWI). Materials and Methods:This retrospective study included 29 patients with ovarian granulosa cell tumors and 61 cases with fibroma-thecomas with high-signal intensity on DWI,which were confirmed by surgical pathology in Peking University Third Hospital from January 2019 to October 2023. All tumors were randomly divided into a training set and a validation set at a ratio of 7:3. The clinical model was constructed by clinical and routine MRI features which were selected by univariate analysis and multivariate logistic regression. Radiomics features were extracted from T2WI. Select K best and least absolute shrinkage and selection operator (LASSO) algorithm were used to reduce the dimension and then the radiomics model was constructed by selected features,and a radiomics score (Rad-score) was calculated. The nomogram model was constructed by combining with clinical model and Rad-score. The receiver operator characteristic (ROC) curves were used to evaluate the performance of three models. The decision curve analysis (DCA) was used to evaluate the clinical value. Results:The logistic regression results showed that a "honeycomb" cyst[odds ratio (OR)=0.20,95％ confidence interval (CI)=0.05-0.79,P=0.022]and intratumoral hemorrhage (OR=0.16,95％ CI=0.03-0.98,P=0.048) can be used to construct the clinical model. A total of 9 features were extracted from T2WI to build the radiomics model. Finally,the nomogram model incorporating Rad-score,a "honeycomb" cyst and intratumoral hemorrhage was established. The AUCs of radiomics model and nomogram model were higher than those of clinical model (training set:0.983 vs. 0.742,Z=−4.058,P<0.001;0.969 vs. 0.742,Z=−3.817,P<0.001. validation set:0.858 vs. 0.731,Z=−1.388,P=0.165;0.883 vs. 0.731,Z=−1.612,P=0.107). There was no significantly difference in AUCs between the radiomics model and nomogram model (training set:Z=−1.040,P=0.298;validation set:Z=0.822,P=0.411). DCA results showed that the nomogram model and radiomics model had higher net benefits than the clinical model. Conclusions:The MRI-based radiomics model and nomogram model constructed in this study can distinguish ovarian granulosa cell tumor from ovarian fibroma-thecoma with high-signal intensity on DWI effectively,which is better than the conventional T2WI-based clinical model.
国家科技期刊平台
NSTL
万方数据
