基于动态增强MRI影像组学评分和激素受体的列线图预测乳腺癌新辅助化疗不敏感的价值

目的 探讨基于动态增强MRI（DCE-MRI）影像组学评分（Radscore）和激素受体状态的列线图预测乳腺癌新辅助化疗（NAC）疗效不敏感的价值。 方法 回顾性收集128例行乳腺癌NAC治疗的女性病人,平均年龄（49.2±10.0）岁。128例病人按照7∶3比例随机分为训练集90例（疗效敏感者47例,疗效不敏感者43例）和测试集38例（疗效敏感者15例,疗效不敏感者23例）。基于DCE-MRI影像提取并筛选影像组学特征,采用多因素逻辑回归构建影像组学模型并计算模型的Radscore。采用t检验、χ²检验或Fisher确切概率检验比较训练集和测试集中临床病理指标年龄、雌激素受体（ER）、孕激素受体（PR）、人表皮生长因子受体-2（HER-2）和肿瘤增殖细胞核抗原-67（Ki-67）],将差异有统计学意义的临床病理指标和Radscore纳入多因素逻辑回归,建立联合模型和列线图。应用受试者操作特征（ROC）曲线下面积（AUC）评价影像组学模型和联合模型的预测效能。应用决策曲线评估影像组学模型和联合模型的临床应用价值。 结果 在训练集中,ER和PR在疗效敏感与不敏感组间的差异均有统计学意义（均P<0.05）,但未得到测试集的验证（均P>0.05）。在训练集中,联合模型预测NAC不敏感的AUC值和准确度分别高于影像组学模型约3.8%和3.1%。在测试集中,联合模型预测NAC不敏感的AUC值高于影像组学模型,其较后者提高了约2.3%,但两者的准确度相同。在基于ER、PR和Radscore构建的联合模型列线图中,Radscore得分最高,其次是ER和PR。决策曲线分析显示联合模型的临床获益高于影像组学模型。 结论 基于DCE-MRI的Radscore和ER、PR构建的联合模型列线图能够较好地预测NAC疗效不敏感。

A nomogram based on DCE-MRI Radscore and hormone receptor status for predicting drug insensitive to neoadjuvant chemotherapy in breast cancer patients

Objective To explore the value of nomogram based on dynamic contrast-enhanced MRI (DCE-MRI) Radscore and hormone receptor status in predicting drug insensitive to neoadjuvant chemotherapy (NAC) in breast cancer (BC) patients. Methods A total of 128 women with BC who had received NAC were retrospectively analyzed, with mean age of 49.2±10.0 years. The 128 patients were randomly divided into a primary dataset (consisting of 90 women, including 47 cases were drug-sensitive and 43 cases were drug-insensitive) and a validation dataset (consisting of 38 women, including 15 cases were drug sensitive and 23 cases were drug insensitive) at a rate of 7∶3. The radiomics features were extracted and selected to develop the radiomics model, and Radscore was calculated for the radiomics model. The t test, chi-squared test, or Fisher test were used to compare the clinicopathologic features including age, estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor-2 (HER-2), and tumor expression of the proliferation antigen-67 (Ki-67)] between the drug-sensitive and -insensitive groups. The Radscore combined with all statistically significant clinicopathologic features were applied to develop the combined model and nomogram by using multivariable logistic regression analysis. The efficiency of the radiomics model and combined model were tested by using the area under the curve (AUC) of receiver operating characteristic curve analysis. Clinical usefulness was evaluated by decision curves. Results The ER and PR were significantly different between drug-sensitive and -insensitive groups in the primary dataset (all P<0.05), which was not confirmed in the validation dataset (all P>0.05). The combined model had higher AUC and accuracy than the radiomics model in the primary dataset, with a AUC and accuracy of 3.8% and 3.1% increase respectively. The combined model had a higher AUC than the radiomics model in the validation dataset, with an AUC of 2.3% increase and the same accuracy. The nomogram based on ER, PR, and Radscore showed that Radscore achieved the highest score, followed by the ER and PR. The decision curve showed that the combined model achieved a better net benefit than the radiomics model. Conclusion The nomogram that incorporated PR, ER and Radscore can be used to preoperatively predict drug insensitive to NAC in BC patients.