改进的深度信念网络肺结节良恶性分类  

Classification method of pulmonary nodules based on improved deep belief network

作　　者：张婷 赵文婷 赵涓涓[1] 强彦[1] ZHANG Ting;ZHAO Wen-ting;ZHAO Juan-juan;QIANG Yan(College of Computer Science and Technology,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学计算机科学与技术学院,山西太原030024

出　　处：《计算机工程与设计》2018年第9期2707-2713,2729,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61373100);虚拟现实技术与系统国家重点实验室开放基金项目(BUAA-VR-17KF-14;BUAA-VR-17KF-15);山西省回国留学人员科研基金项目(2016-038)

摘　　要：为解决传统的计算机辅助诊断系统中肺结节特征提取过程复杂的问题,提出一种基于深度信念网络的肺结节良恶性分类方法。通过阈值概率图对原始CT图像进行预处理,采用多隐层深度信念网络提取肺结节图像的深层特征,引入交叉熵稀疏惩罚因子机制解决受限玻尔兹曼机在训练过程中出现的特征同质化现象,将极限学习机作为分类器对提取到的特征进行良恶性分类。通过对比多种深度学习方法在肺结节诊断方面的优势与不足,验证了该方法的准确性、特异性、敏感性均优于其它算法。To solve the complex problem of the feature extraction of pulmonary nodules in traditional computer-aided diagnosis system,a method for classifying benign and malignant pulmonary nodules based on deep belief network was proposed.The original CT image was preprocessed using the threshold probability maps method,and the cross-entropy sparse penalty factor was used to prevent the phenomenon of pulmonary nodule homogeneity in training processes of restricted Boltzmann machine.The deep features of the pulmonary nodules were extracted using the multi-hidden deep belief network.An extreme learning machine was used as the classifier for benign and malignant classification.Comparing the advantages and disadvantages of other deep learning methods in the diagnosis of pulmonary nodules,the proposed method is superior to other algorithms in terms of accuracy,specificity and sensitivity.

关 键 词：肺结节 深度信念网络 特征提取 极限学习机 良恶性分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]