超声影像组学对致密型乳腺背景中非肿块型乳腺癌的诊断价值  

Diagnostic value of ultrasound radiomics in non-mass breast cancer in dense breasts

作　　者：邱琳 刘锦辉 组木热提·吐尔洪 马悦心 冷晓玲 Lin Qiu;Jinhui Liu;Tuerhong Zumureti;Yuexin Ma;Xiaoling Leng(Department of Ultrasound,Affiliated Tumor Hospital of Xinjiang Medical University,Urumqi 830011,China;Department of Ultrasound,Tenth Affiliated Hospital of Southern Medical University/Dongguan People’s Hospital,Dongguan 523059,China)

机构地区：[1]新疆医科大学附属肿瘤医院超声诊断科,乌鲁木齐830011 [2]南方医科大学附属第十医院(东莞市人民医院)超声科,东莞523059

出　　处：《中华乳腺病杂志(电子版)》2024年第6期353-360,共8页Chinese Journal of Breast Disease(Electronic Edition)

摘　　要：目的探讨超声影像组学特征对检出致密型乳腺背景中的非肿块型乳腺癌的诊断价值。方法回顾性分析2017年1月1日到2023年1月30日东莞市人民医院及新疆医科大学附属肿瘤医院619例致密型乳腺背景中的非肿块型病变(NML)的二维超声图像,采用7∶3的比例进行随机分组,训练组434例,验证组185例,共提取848个影像组学特征,采用最小绝对收缩和选择算子(LASSO)回归模型进行特征筛选,通过LASSO-Logistic回归来建立影像组学模型,并与临床和超声特征进行整合构建联合模型。通过比较受试者工作特征(ROC)曲线评价模型的诊断效能。用校准曲线评估模型的一致性,用决策曲线分析(DCA)评估模型的临床价值,用DeLong检验将其与其余模型进行比较。结果术后病理结果显示619例乳腺NML中,恶性304例,良性315例。单因素和多因素Logistic回归分析结果显示,年龄、病灶长度、细小钙化、周围结构扭曲、血流特征是恶性病变的独立预测因素(OR=1.053、8.197、0.701、3.479、1.195;95%CI:1.027~1.080,4.895~14.154,0.573~0.857,2.044~6.044,1.536~2.408;P均<0.050)。共筛选出12个非零系数的影像组学特征。将筛选出临床指标和影像组学特征整合,创建了联合预测模型。联合预测模型的训练组ROC曲线下面积为0.89(95%CI:0.86~0.92),验证组曲线下面积为0.83(95%CI:0.78~0.89)。DeLong检验表明,联合模型与临床模型、超声模型、影像组学模型比较,差异有统计学意义(Z=-3.974、-3.338、-3.468,P均<0.050)。联合模型的DCA曲线下面积最大,训练组为0.12,验证组为0.22。校准曲线显示,与其他模型相比,联合模型在预测结果与真实病理结果具有更好的一致性。结论超声影像组学与临床指标的联合模型对于致密型乳腺背景中NML的良恶性的鉴别具有较好的效能,可为乳腺癌的临床治疗决策提供支持。Objective To investigate the diagnostic value of ultrasound radiomics features for the detection of non-mass breast cancer in dense breasts.Methods We retrospectively analyzed 2D ultrasound images of 619 patients with non-mass breast lesions(NML)in dense breasts in Dongguan People’s Hospital and Affiliated Tumor Hospital of Xinjiang Medical University between January 1st,2017 and January 30th,2023.They were randomized into two groups using a 7∶3 ratio(434 cases in the training group and 185 cases in the validation group).Totally 848 imaging features were extracted.The least absolute shrinkage and selection operator(LASSO)regression model was used to screen the features,and the radiomics model was built by the LASSO-Logistic regression.The joint model was constructed by integrating it with clinical and ultrasound features.The diagnostic efficacy of the model was evaluated by the receiver operating characteristic(ROC)curve.The consistency of the model was assessed with the calibration curve and the clinical value of the model was assessed by decision curve analysis(DCA).The DeLong test was used to compare those models.Results The postoperative pathological results showed that among 619 cases of breast NML,304 cases were malignant and 315 cases were benign.The results of univariate and multivariate Logistic regression analyses indicated that age,lesion length,microcalcification,distortion of surrounding structures,and blood flow features were independent predictors of malignancy(OR=1.053,8.197,0.701,3.479,1.195;95%CI:1.027-1.080,4.895-14.154,0.573-0.857,2.044-6.044,1.536-2.408;all P<0.050).A total of 12 radiomics features with non-zero coefficients were screened out.The selected clinical factors and radiomics features were integrated to create a joint prediction model.The area under the ROC curve of the joint model was 0.89(95%CI:0.86-0.92)in the training group,and 0.83(95%CI:0.78-0.89)in the validation group.The area under the DCA curve of the joint model was the largest(0.12 in the training group and 0.22 in t

关 键 词：乳腺肿瘤 影像组学 致密型乳腺 非肿块型乳腺病变 

分 类 号：R737.9[医药卫生—肿瘤]