基于遗传算法和BP神经网络的孤立性肺结节分类算法

Solitary Pulmonary Nodules Classification Based on Genetic Algorithm and Back Propagation Neural Networks

为了提高计算机辅助诊断系统中孤立性肺结节的良恶性诊断的准确性,提出了一种基于遗传算法和BP神经网的分类算法。该算法针对BP神经网络容易陷入局部最优的问题,综合考虑孤立性肺结节的医学诊断特性,采用遗传算法对基于BP神经网络的分类器进行优化,并通过对PET/CT图像进行处理,提取病灶的功能特征、结构特征以及临床信息作为神经网络分类器的输入样本,实现孤立性肺结节的良恶性分类。对医院以及网络公共数据库中的大量实验数据进行分类实验,结果表明优化后的算法在分类准确性上有较大的提高,说明该方法在肺结节临床分类方面是有效的。

In order to improve the accuracy of benign and malignant diagnosis of the solitary pulmonary nodules in the computer aided diagnosis system,this paper proposed a novel classification algorithm based on genetic algorithm and back propagation neural networks.Considering the local optimum problem of the BP neural networks and the medical diagnosis features of solitary pulmonary nodules,the proposed algorithm uses genetic algorithm to optimize the classifier based on BP neural networks.Through the PET/CT image processing,the functional characteristics,structural characteristics and clinical information of the lesions are extracted as input samples of the neural network based classifier.Then,the benign and malignant diagnosis of the solitary pulmonary nodules is realized by the novel classifier.Classify experimental results on a large number of experiment data from a hospital and public databases on network show that the optimized algorithm is greatly improved on the classification accuracy,indicating that this method is effective in clinical classification of pulmonary nodules.