基于深度卷积神经网络的乳腺钼靶筛查

Mammary Molybdenum Target Screening Based on Deep Convolutional Neural Network

乳腺钼靶X线摄影检查是乳腺癌的筛查与诊断的重要手段之一,临床上应用广泛。随着计算机技术的发展,通过计算机算法实现对乳腺钼靶图像的自动诊断可以有效缓解医疗压力,提高诊断精度。在此基础上,通过对乳腺钼靶图像的分析,提出了一种基于深度学习的乳腺钼靶图像自动诊断方法。首先通过采样制作的切块数据集训练了一个识别乳腺钼靶中良恶性肿块及钙化病灶的多分类网络,之后使用训练好的网络实现对整张图像的遍历识别,最终返回良恶性肿块及钙化的预测概率热图。结果表明,本文算法对于切块数据的识别精度可以达到93%左右,生成的预测概率热图可以对医生提供较大的帮助。

Mammography X-ray radiography examination is one of the most important methods for breast cancer screening and early diagnosis,which is widely used in clinic application.With the development of computer technology,automatic diagnosis of mammography images can effectively relieve medical pressure and improve diagnostic accuracy.Based on that,we propose an automatic diagnosis method of mammography based on deep learning algorithm after analyzing mammography images.Firstly,a multi-classification network is trained to recognize benign and malignant tumors and calcified lesions in mammogram target by sampling data sets.Then,the trained network is used to recognize the whole image traversally,and finally the predictive probability heat map of benign and malignant tumors and calcification is returned.The results show that the accuracy of the algorithm for patches dataset can reach about 93%,and the predicted probability heatmaps can be of great help to doctors.