改进Xception模型的乳腺钼靶图像识别研究

乳腺X线摄影技术是早期发现乳腺癌的主要方法，但其结果很大程度上受放射科医师临床诊断经验的限制；基于卷积神经网络对乳腺钼靶图像自动分类的研究可以为放射科医师临床诊断提供意见，然而乳腺癌肿块边缘模糊且良恶性肿块特征差异较小，分类任务面临重重挑战；为了提高乳腺钼靶图像分类的准确率，提出一种基于Xception模型的改进优化算法，改进模型中的残差连接模块，并嵌入Squeeze-and-excitation（SE）注意力机制对模型进行优化；采用优化后的Xception模型并结合迁移学习算法进行乳腺钼靶图像特征提取，并优化全连接层网络进行图像分类，使用公开的乳腺癌图像数据库CBIS-DDSM进行实验，将乳腺钼靶图像自动分为良性和恶性；实验结果表明该方法可以有效提高模型的分类效果，准确率和AUC分别达到了97.46%和99.12%。

Research on Mammograms Recognition With Improved Xception Model

Mammography is the primary method for the early detection of breast cancer, but the results is largely limited by the radiologist''s experience in clinical diagnosis. The study of automatic classification of mammography images based on convolutional neural network can provide advice for radiologists in clinical diagnosis, however the classification task of mammography images faced with many challenges due to the fuzzy edge and small difference between benign and malignant tumors. In order to improve the accuracy of mammography classification, an improved optimization algorithm based on Xception model was proposed, the residual connection module in the model is improved, and Squeeze-and-excitation(SE) attention mechanism is embedded to optimize the model. The optimized Xception model combined with transfer learning algorithm was used to feature extraction of mammography images, and the full-connection layer network was optimized for image classification. Experiments were conducted on the open data set CBIS-DDSM, and mammography images were automatically divided into benign and malignant. The experimental results showed that this method could effectively improve the classification effect of the model, and the accuracy and AUC reached 97.46% and 99.12%, respectively.