基于ADC影像组学的机器学习模型预测子宫内膜癌肌层浸润深度的价值  

Predictive value of machine learning model based on ADC radiomics in evaluating the invasion depth of endometrial carcinoma

作　　者：崔靖 郭冉 信瑞强[1] CUI Jing;GUO Ran;XIN Ruiqiang(Department of Radiology,Beijing Luhe Hospital,Captital Medical University,Beijing 101199,China)

机构地区：[1]首都医科大学附属北京潞河医院放射科,北京101199

出　　处：《磁共振成像》2025年第3期77-82,共6页Chinese Journal of Magnetic Resonance Imaging

摘　　要：目的 探讨基于表观扩散系数(apparentdiffusioncoefficient,ADC)图构建的影像组学模型,对子宫内膜癌(endometrial carcinoma,EC)肌层浸润深度的预测价值,从而为临床制订治疗方案提供可靠依据。材料与方法 回顾性分析首都医科大学附属北京潞河医院2016年1月至2023年12月期间在术前接受盆腔MRI检查并经术后病理证实的155例EC患者的临床及MRI资料(浅肌层浸润114例,深肌层浸润41例),按照4∶1的比例随机分为训练集(n=124)和验证集(n=31)。采用ITK-SNAP软件在ADC图上逐层勾画肿瘤区域并进行特征提取,对提取出来的特征进行归一化处理,应用皮尔森相关系数分析(pearson correlation coefficients,PCC)及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)对所有特征进行筛选降维,并按权重系数对筛选后的影像组学特征进行重要性排序,选择排名前10的特征,使用逻辑回归(logistic regression,LR)、随机森林(random forest,RF)、梯度提升机(gradient boosting machine,GBM)3种算法构建影像组学模型,并在验证集中对模型进行验证。使用受试者工作特征(receiver operating characteristic,ROC)曲线、校准曲线和决策曲线分析(decision curve analysis,DCA)对3种影像组学模型的性能进行分析评估。使用DeLong检验比较不同模型间曲线下面积(area under the curve,AUC)的差异。结果 LR、RF和GBM模型预测EC肌层浸润深度的AUC值分别是0.780 (95%CI:0.762~0.804)、0.860(95%CI:0.846~0.879)、0.860(95%CI:0.843~0.877),RF和GBM模型的AUC值最高且相等。DeLong检验显示LR与RF及GBM模型的AUC值差异均有统计学意义(P=0.017,0.023),RF与GBM模型的AUC值差异无统计学意义(P=3.310)。校准曲线和DCA结果显示3种模型均具有较好的拟合度及临床实用性。结论 基于ADC图所构建的影像组学模型在EC肌层浸润深度的预测中具有良好的价值。Objective:To explore the predictive value of radiomics models based on apparent diffusion coefficient(ADC) in evaluating the myometrial invasion depth of endometrial carcinoma(EC),providing a reliable evidence for clinicians to formulate treatment plans.Materials and Methods:Retrospective analysis of 155 patients with EC who underwent preoperative pelvic MR examination and were confirmed by pathology after operation from January 2016 to December 2023 in Beijing Luhe Hospital(superficial myometrial invasion = 114,deep invasion = 41),and randomly divided into training set(n = 124) and validation set(n = 31) in a 4∶ 1 ratio.The ITK-SNAP software was used to delineate the tumor regions layer by layer on the ADC maps,and the radiomics features were extracted,the extracted features were normalized.Pearson correlation coefficients(PCC) and least absolute shrinkage and selection operator(LASSO) were used to reduce features dimensionality,and the importance of the screened radiomics features was ranked according to the weight coefficient,the top 10 features were used to build radiomics models using three algorithms:logistic regression(LR),random forest(RF),and gradient boosting machine(GBM).The models were validated on the validation set.The performance of three radiomics models were evaluated by the receiver operating characteristic(ROC) curve,calibration curves,and decision curve analysis(DCA).The AUC values were compared using the DeLong test.Results:The AUC values of the LR,RF,and GBM models in predicting the invasion depth of endometrial carcinoma were 0.780(95% CI:0.762 to 0.804),0.860(95% CI:0.846 to 0.879),and 0.860(95% CI:0.843 to 0.877),respectively.The AUC values of the RF and GBM were the highest and equal.The DeLong test showed that there was a statistically significant difference in AUC values between LR,RF,and GBM models(P = 0.017,0.023),while there was no statistically significant difference in AUC values between RF and GBM models(P = 3.310).The calibration curve and DCA curve show that all three models

关 键 词：子宫内膜肿瘤 肌层浸润 磁共振成像 影像组学 机器学习 表观扩散系数 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.33[医药卫生—诊断学]