基于改进的卷积神经网络的肺结节良恶性分类

为提高肺结节良恶性分类的准确率,降低误诊率,提出一种基于改进的卷积神经网络的肺结节良恶性分类方法。采用多层感知器卷积层来提取肺结节特征;利用卷积层代替全连接层,减少网络参数,将提取到的特征输入至分类器进行分类;从网络深度、参数优化算法、学习率衰减策略、激活函数4个方面分析对分类效果的影响,构建改进的卷积神经网络模型。在LIDC-IDRI数据集上的实验结果表明,该模型的准确率、敏感性、特异性和AUC值分别为95.5%、0.96、0.95和0.96,该方法比传统卷积神经网络有更高的分类准确率和低误诊率,取得了较好的分类效果。

Classification of benign and malignant pulmonary nodules based on improved convolution neural network

To improve the accuracy of classification of benign and malignant pulmonary nodules,and reduce the rate of misdiagnosis,a classification method based on improved convolution neural network for benign and malignant pulmonary nodules was proposed. The multi-layer perceptron convolution layer was used to extract pulmonary nodules. The fully connected layer was replaced using the convolution layer to reduce network parameters,and the extracted features were inputted to the classifier for classification. From four aspects of the network depth,the parameter optimization algorithm,the learning rate attenuation stra- tegy and the activation function,the influence of the classification effect was analyzed,and the improved convolution neural network model was constructed. Experimental results on the LIDC-IDRI dataset show that the accuracy,sensitivity,specificity and AUC of the model are 95.5%,0.96,0.95 and 0.96 respectively. The proposed method has higher classification accuracy and lower misdiagnosis rate than traditional convolution neural network,and achieves better classification results.