【摘要】 目的 本研究旨在探讨深度学习超声组学鉴别肿块型乳腺炎(MM)和浸润性乳腺癌(IBC)的应用价值。方法 回顾性分析经病理证实的50例MM和180例IBC的超声图像。基于ResNet50和GoogLeNet深度学习神经网络提取深度学习特征，并采用3种分类器构建深度学习模型。使用受试者工作特征(ROC)曲线评估模型的效果。结果 基于两种深度学习神经网络的特征均表现为具有中度以上鉴别性能，且ResNet50特征效果更优。基于多种不同分类器构建的深度学习模型均可有效鉴别MM和IBC(AUC≥0.75)。结论 深度学习超声组学有助于鉴别MM与IBC,具有转化为计算机辅助系统助力于大规模乳腺疾病筛查的潜力，提高基层医院的乳腺疾病超声诊断水平，更有利于患者的个体化精准诊疗。更多还原

【Abstract】 Objective This study aims to explore the application value of deep learning ultrasound-based radiomics in differentiating mass mastitis(MM) and invasive breast cancer(IBC). Methods Ultrasound images of 50 cases of MM and 180 cases of IBC confirmed by pathology were retrospectively analyzed. Based on ResNet50 and GoogLeNet deep learning neural networks, deep learning features were extracted and three classifiers were used to construct deep learning models. The receiver operating characteristic(ROC) curve was used to evaluate the performance of the model. Results The features based on two types of deep learning neural networks exhibit moderate or higher discriminative performance, and ResNet50 features perform better. Deep learning models based on a variety of different classifiers could effectively distinguish MM and IBC(AUC≥0.75). Conclusions Deep learning ultrasound-based radiomics is helpful in differentiating MM from IBC, and has a potential value to be transformed into a computer-aided system to large-scale breast disease screening, which may improve ultrasound diagnosis of breast disease for primary care hospitals, and be more conducive in individualized and precise treatment for patients.更多还原