基于深度信念网络的肺结节良恶性分类

肺结节的良恶性分类是计算机辅助诊断系统中最重要的部分,目前常用的分类方法有分类精度低、假阳性高等问题。针对上述问题,把深度信念网络(DBN)引入肺结节的良恶性诊断过程中,提出自定义的DBN分类算法。首先从不同的角度提取肺结节特征,并形成特征向量。然后根据提取的特征对三个隐藏层的节点数进行分析;并构建了一个5层深度信念网络。最终使用训练样本对DBN进行训练;并输出网络的测试结果。对175个病例进行试验,结果表明:算法的分类精度、敏感性和特异性分别为95.3%,92.5%和93.2%,ROC曲线下面积为0.921。与传统算法相比有更好的分类效果,可以给医生提供客观的辅助诊断。

A Classification Method of Pulmonary Nodules Based on Deep Belief Network

The classification of benign and malignant lung nodules is the most important part of the computer aided diagnosis system. Commonly used classification methods have low classification accuracy and high false positive rate currently. Aiming atSabove problems, we use deep belief network (DBN) in the benign and malignant diagnosis process of pulmonary nodules and propose a classification algorithm based on DBN with self-define structure. Firstly, the characteristics of pulmonary nodules are extracted from different aspects and the feature vectors are formed. Then analyzing the number of nodes in three hidden layers according to extracted features and constructing a deep belief network with 5 layers. Finally, training and testing the classifier based on deep belief network. The test results of 175 cases show that the classification accuracy, sensitivity and specificity of the method proposed in this paper are 95.3%, 92.5% and 89.2% respectively