利用MRI影像组学模型可有效预测乳腺癌前哨淋巴结转移

目的 探索基于不同乳腺MRI序列联合临床病理因素的MRI影像组学模型预测乳腺癌前哨淋巴结转移的潜在价值。 方法 回顾性分析182例经病理确诊为乳腺癌伴前哨淋巴结转移状态患者，其中前哨淋巴结转移阳性组91例，前哨淋巴结转移阴性组91例，并按7：3的比例分训练组（阳性组64例、阴性组64例）和验证组（阳性组27例、阴性组27例）。对乳腺癌患者的临床、影像及病理资料进行单因素和多因素Logistic回归分析并筛选出与乳腺癌前哨淋巴结转移相关的独立风险因素；基于T2WI、弥散加权成像、动态对比增强提取最佳影像组学特征，分别构建多个单、多序列影像组学标签评分，并结合临床、病理及影像特征独立风险因素建立影像组学联合预测模型。绘制ROC曲线，计算曲线下面积，评价各模型预测乳腺癌前哨淋巴结转移的效能。 结果 瘤周水肿（P < 0.001）、肿瘤长径（P < 0.001）、肿瘤短径（P < 0.001）、病理分级（P < 0.001）、脉管侵犯（P < 0.001）、毛刺征（P= 0.006）、弥散加权成像边缘高信号征（P=0.028）及表观弥散系数值（P < 0.001）为乳腺癌前哨淋巴结转移的独立临床病理因素。在影像组学标签评分中，T2WI+弥散加权成像+动态对比增强联合序列的影像组学标签评分预测效能最佳，其验证组曲线下面积为0.744，进一步联合临床、病理及影像特征独立风险因素建立的影像组学联合预测模型的预测效能得到进一步提高，其验证组曲线下面积为0.834。 结论 基于乳腺MRI的影像组学模型在术前能够有效预测乳腺癌前哨淋巴结的转移。

Breast MRI-based imaging radiomic model can effectively predict sentinel lymph node metastasis in breast cancer prior to surgery

Objective To explore the potential value of radiomics model based on different MRI sequences of breast combined with clinicopathological factors in predicting sentinel lymph node metastasis of breast cancer. Methods We retrospectively analyzed 182 cases of breast cancer with sentinel lymph node metastasis diagnosed by pathology, including 91 in the sentinel lymph node positive group and 91 in the sentinel lymph node negative group, and divided them into a training group (64 positive and 64 negative) and a validation group (27 positive and 27 negative) according to the ratio of 7:3. The clinical, imaging and pathological data of breast cancer patients were analyzed by univariate and multivariate logistic regression, and the independent risk factors related to sentinel lymph node metastasis of breast cancer were screened out. Based on T2WI, diffusion-weighted imaging and dynamic contrast enhancement, the best imaging features were extracted, and several singlesequence and multi- sequence radiomics label scores were constructed respectively, and the combined radiomics prediction model was constructed combined with the above independent risk factors of clinical, pathological and imaging features. The effectiveness of each model in predicting breast cancer sentinel lymph node metastasis was evaluated by plotting the ROC curves and calculating the area under the curve (AUC). Results Peritumoural edema (P < 0.001), tumour long diameter (P < 0.001), tumour short diameter (P < 0.001), pathological grade (P < 0.001) and vascular infiltration (P < 0.001), burr sign (P=0.006), diffusion-weighted imaging rim high signal sign (P=0.028) and ADC value (P < 0.001) were the independent clinicopathological factors of anterior sentinel lymph node metastasis in breast cancer. Among the radiomics label scores, the multi- sequence radiomics label score of T2WI+ diffusion-weighted imaging+dynamic contrast enhancement had the best predictive efficiency, its AUC in the validation group was 0.744, and the predictive efficiency of the combined radiomics prediction model established by combining clinical, pathological and imaging feature independent risk factors had been further improved, and its AUC in the validation group was 0.834. Conclusion The breast MRI-based imaging radiomic model can effectively predict sentinel lymph node metastasis in breast cancer prior to surgery