全视野数字化乳腺X线摄影影像组学及深度学习特征预测乳腺癌HER-2状态

Study on predicting breast cancer HER-2 status through full-field digital mammography radiomic and deep learning features

目的通过联合全视野数字化乳腺X线摄影(FFDM)影像组学特征及深度学习特征预测乳腺癌HER-2状态。方法回顾性分析山西中医药大学附属医院2018年3月~2023年12月经临床手术或穿刺活检的乳腺癌患者FFDM、临床及病理资料。FFDM图像手工勾画肿瘤感兴趣区并提取组学特征和深度学习特征,分别经过LASSO特征筛选后采用支持向量机算法建立影像组学模型和深度学习模型,通过多因素逻辑回归分析建立综合模型。计算各模型曲线下面积(AUC)评估其预测效能,并通过决策曲线分析评估各模型在实际临床决策中的有效性和实用价值。结果影像组学模型在训练集和测试集的AUC分别为0.835(95%CI:0.761~0.898)和0.781(95%CI:0.701~0.857),深度学习模型在训练集和测试集的AUC分别为0.904(95%CI:0.855~0.955)和0.883(95%CI:0.823~0.939),综合模型在在训练集和测试集的AUC分别为0.921(95%CI:0.872~0.967)和0.890(95%CI:0.828~0.942)。决策曲线分析显示3种模型相比于全部认为HRE-2阳性或阴性可获得更好的净收益,其中综合模型在风险阈值下可获得最大净收益。结论基于FFDM影像组学特征和深度学习特征的联合应用,可以更有效地预测乳腺癌HER-2状态,显著提高了预测的准确性和可靠性,为乳腺癌的诊断和治疗开辟了新的途径。

Objective To predict the HER-2 status in breast cancer by integrating full-field digital mammography(FFDM)radiomics features with deep learning features.Methods Retrospective analysis was conducted on of breast cancer patients who underwent clinical surgery or biopsy from March 2018 to December 2023 at the Affiliated Hospital of Shanxi University of Chinese Medicine.Regions of interest within FFDM images were manually delineated to extract radiomics and deep learning features.Following LASSO-based feature selection,support vector machine algorithms were used to construct both a radiomics model and a deep learning model.A composite model was developed using multivariate logistic regression analysis.The predictive performance of each model was evaluated by calculating their AUC values,and their effectiveness and practical value in real clinical decision-making were assessed using decision curve analysis curves.Results The radiomics model exhibited AUC values of 0.835(95%CI:0.761-0.898)in the training set and 0.781(95%CI:0.701-0.857)in the test set.The deep learning model demonstrated AUC values of 0.904(95%CI:0.855-0.955)in the training set and 0.883(95%CI:0.823-0.939)in the test set.The composite model achieved AUC values of 0.921(95%CI:0.872-0.967)in the training set and 0.890(95%CI:0.828-0.942)in the test set.Decision curve analysis indicated that all three models provided a greater net benefit compared to assuming all cases as either HER-2 positive or negative,with the composite model offering the maximum net benefit at various risk thresholds.Conclusion The integration of FFDM radiomics and deep learning features significantly enhances the prediction of HER-2 status in breast cancer,greatly improving both accuracy and reliability.This advancement opens new avenues for the diagnosis and treatment of breast cancer.