基于改进自生成神经网络的孤立性肺结节分类

摘要

为了提高孤立性肺结节良恶性诊断中的分类准确度，提出了一个基于自生成神经网络的自动分类算法。该算法首先对PET／CT 图像进行去噪、配准等预处理，分别提取孤立性肺结节的结构影像特征和代谢特征，然后对自生成神经网络进行训练和优化，构建分类器，根据距离测度和自动连接规则对待分类肺结节进行分类。初步的实验结果表明，与传统的自生成神经网络算法和BP神经网络算法相比，改进的自生成神经网络分类算法能得到更高的分类准确率。%To improve the classification accuracy of diagnosis of solitary pulmonary nodules , this paper proposes an automatic classification algorithm based on the self‐generating neural net‐works .The algorithm first deploys preprocessing methods on PET/CT image ,such as denoising and registration .Then ,structural and metabolic features of solitary pulmonary nodules are extrac‐ted separately .After that ,the improved self‐generating neural network is trained and a classifier is built to classify unspecified samples on the basis of distance measure and automatic connection . Experimental results show that compared with traditional self‐generating neural networks and BP neural network algorithm ,the improved self‐generating neural networks algorithm can guarantee higher classification accuracy .