基于超声视频的深度学习在乳腺结节良恶性鉴别诊断中的价值

Value of Deep Learning Based on Ultrasonic Video in the Differential Diagnosis of Benign and Malignant Breast Nodules

目的:探讨基于U-net和PyRadiomics的一种深度学习与影像组学相结合的方法在乳腺结节良恶性鉴别诊断中的应用价值。方法:纳入上海市静安区中心医院285例乳腺结节患者超声影像视频资料,其中乳腺良性结节216例,乳腺恶性结节69例。经图像预处理和U-net图像分割后,通过PyRadiomics提取图像病灶特征,经过特征选择降维得到17种特征,使用逻辑回归算法对乳腺结节良恶性进行分类,绘制受试者工作特征曲线和混淆矩阵分析深度学习后模型对乳腺良恶性结节的诊断效能。结果:基于超声视频的深度神经网络模型对乳腺良恶性诊断的准确度、灵敏度、特异度、曲线下面积分别为86.0%、85.7%、86.0%、0.890。结论:基于超声视频的深度神经网络模型在乳腺结节良恶性鉴别诊断中具有较高的应用价值。

Purpose:To explore the application value of deep neural network multiple instance learning model based on ultrasonic video in the differential diagnosis of benign and malignant breast nodules.Methods:The ultrasonographic video data of 285 patients with breast nodules in Jing'an District Central Hospital of Shanghai were collected,including 216 benign nodules and 69 malignant nodules.After image preprocessing and U-net semantic segmentation,the focal characteristics were extracted by PyRadiomics,and 17 features were obtained by feature selection and dimensionality reduction.Logistic regression algorithm was used to classify benign and malignant breast nodules.Receiver operating characteristic curve and confusion matrix were used to analyze the diagnostic efficacy of the deep-learning model for benign and malignant breast nodules.Results:The accuracy,sensitivity,specificity and area under curve of the deep neural network model for benign and malignant breast diagnosis were 86.0%,85.7%,86.0%and 0.890,respectively.Conclusion:The deep neural network model based on ultrasonic video is with high application value in the differential diagnosis of benign and malignant breast nodules.