基于瘤内与瘤周乳腺断层合成X线摄影影像组学列线图可有效预测乳腺癌腋窝淋巴结转移

Value of predicting axillary lymph node metastasis of breast cancer based on intra tumoral and peri-tumoral digital breast tomosynthesis imaging Nomogram

目的探讨基于乳腺断层合成X线摄影(DBT)的瘤内联合瘤周影像组学特征及临床因素的列线图在乳腺癌腋窝淋巴结(ALN)转移术前预测中的价值。方法回顾性收集2019年1月~2023年12月于蚌埠医科大学第一附属医院行乳腺DBT检查的210例女性乳腺癌患者资料,以7:3比例随机分为训练集(n=147)与验证集(n=63)。选择DBT图像肿块最大层面,医师手动勾画瘤内感兴趣区域,自动外扩3 mm获取瘤周感兴趣区域;提取并筛选影像组学特征,利用支持向量机构建瘤内、瘤周、瘤内+瘤周组学模型并计算预测值;选择效能最高的组学模型预测值联合临床特征构建列线图模型。以ROC曲线、校准曲线及决策曲线评估模型预测性能。结果与单一瘤内、瘤周模型相比,由15个最优组学特征构建的瘤内+瘤周组学模型诊断性能最优。ALN触诊、DBT_ALN及瘤内+瘤周模型预测值为独立危险因素(P<0.05),所构建的列线图模型敏感度、特异度、准确度、曲线下面积在训练集中分别为82.7%、94.7%、86.4%、0.942,在验证集中分别为90.5%、83.3%、87.3%、0.932,实现了最佳预测效能。结论基于DBT瘤内联合瘤周影像组学特征及临床因素的列线图可于术前有效预测乳腺癌ALN转移,可作为一种无创预测方法指导临床决策。

Objective To evaluate the worth of intra-tumoral and peri-tumoral radiomics Nomogram in the pre-operative prognostication of axillary lymph node(ALN)metastasis based on digital breast tomosynthesis(DBT)for breast cancer.Methods A total of 210 breast cancer patients performed breast DBT examinations were retrospectively collected at the First Affiliated Hospital of Bengbu Medical University from January 2019 to December 2023,all patients were stochastically allocated to a training set(n=147)and a verification set(n=63)in a 7:3 ratio.Select the largest dimension of the tumor in the DBT image to manually delineate the ROI of the intra-tumoral region of interest,and the peri-tumoral ROI was obtained by expanding outward by 3 mm.Radiomics features were extracted and screened.Support vector machine was used to construct the models of intra-tumoral,peritumoral and intra-tumoral+peri-tumoral and calculate predictions.The predicted value of the radiomics model with the highest predictive efficiency was selected,and a Nomogram model was created by combining the clinical features.The forecast power of the model was analyzed using the ROC curve,calibration and decision curves.Results The"intra-tumoral+peri-tumoral"model constructed from the 15 best radiomics features performed better than the"intra-tumoral"and"peri-tumoral"models.ALN palpation,DBT_ALN and"intra-tumoral+peri-tumoral"model's forecast value are separate risk elements(P<0.05),and the best predictive efficacy was achieved by the constructed Nomogram model,with sensitivity of 82.7%,specificity of 94.7%,accuracy of 86.4%,AUC of 0.942 in the training set,and 0.932,90.5%,83.3%and 87.3%in the verification set.Conclusion The Nomogram incorporating intra-tumoural and peri-tumoural DBT radiomics characteristics and clinical elements are effective in predicting ALN metastasis before the operation of breast cancer,regarding as a noninvasive predictive approach to assist clinical policy development.