北京生物医学工程

基于Gabor小波和深度置信网络的肺结节良恶性分类研究

Study on the classification of benign and malignant pulmonary nodules based on Gabor wavelet and deep belief network

作者：王亚娟管建王立功

单位：苏州大学医学部放射医学与防护学院（江苏苏州 215123）南京医科大学附属苏州医院（苏州市立医院）放疗科室（江苏苏州 215100）

关键词：孤立性肺结节； Gabor小波；纹理特征；受限玻尔兹曼机；深度置信网络

分类号：R318.04

出版年·卷·期（页码）：2019·38·3（263-270）

摘要：

目的从频率域角度研究孤立性肺结节纹理特征，探讨深度置信网络对其良恶性的分类效果，达到辅助医生提高早期肺癌诊断准确率的目的。方法首先，利用Gabor小波对1012例患者的1072张孤立性肺结节CT图像提取纹理特征，用受限玻尔兹曼机对特征向量进行编码，学习数据本质特征。然后，用得到的纹理特征向量集训练深度置信网络，构建分类模型。最后，通过K折交叉验证法从准确性、ROC曲线下面积（AUC值）以及时间成本方面对本文提出的研究方法进行评估。结果经Gabor小波变换并构建DBN分类模型的准确度为83.75%，测试集的AUC值为0.78。与传统支持向量机分类模型相比，所提方法的准确度上升了0.56%，时间成本缩减了一半。结论利用Gabor小波从频率域提取纹理特征，结合深度置信网络构建分类模型能够取得较好的分类效果，一定程度上能够为临床诊断肺结节的良恶性提供参考。

Objective To explore the texture features of solitary pulmonary nodules in the frequency domain and evaluate their benign and malignant classification effect by virtue of the deep belief networks so as to improve the diagnostic accuracy of the early-stage lung cancer. Methods First, the Gabor wavelet was used to extract the texture features of 1072 solitary pulmonary nodule CT images from 1012 patients in this study. The feature vectors were coded to learn the essential feature of data by the restricted Boltzmann machine. Then the texture feature vector set was used to train the deep belief network and construct the classification model. Finally, the proposed method in this research study was evaluated in terms of accuracy, and the area under the receiver operator characteristic curve (AUC), and time cost through K-fold cross-validation. Results The accuracy for the classification model using the Gabor wavelet and DBN was 83.57%. The AUC value obtained from the testing set was 0.78. Compared to the conventional support vector machine, the accuracy of the proposed method was increased by 0.56% and the time cost was reduced by half. Conclusions Using the proposed Gabor wavelet method to extract the texture features in the frequency domain and constructing the classification model combined with the deep belief network can obtain relatively excellent accuracy, and it to certain extent can offer a reference for the clinical diagnosis on the benign and malignant pulmonary nodules.

参考文献：

[1] 刘春全, 崔永. 肺结节评估四大指南比较分析[J]. 中国肺癌杂志, 2017, 20(7):490-498.

Liu CQ, Cui Y. Lung nodules assessment—analysis of four guidelines[J]. Chinese Journal of Lung Cancer, 2017, 20(7):490-498.

[2] Jemal A. Cancer statistics[J]. CA: A Cancer Journal for Clinicians, 2013, 52(5):1-24.

[3] 张敏鸣. 孤立性肺结节影像学诊断原则与研究进展[J]. 放射学实践, 2007, 22(3):225-229.

[4] 张刚, 马宗民. 一种采用Gabor小波的纹理特征提取方法[J]. 中国图象图形学报, 2010, 15(2):247-254.

Zhang G, Ma ZM. An approach of using gabor wavelets for texture feature extraction[J]. Journal of Image and Graphics, 2010, 15(2):247-254.

[5] 王瓛, 郭秀花, 李坤成,等. 良恶性肺小结节CT图像基于灰度共生矩阵10种纹理特征研究[J]. 北京生物医学工程, 2008, 27(6):561-564.

Wang H, Guo XH, Li KC, et al. CT image 10 texture features of small solitary pulmonary nodules patients using gray level co-occurrence matrix[J]. Beijing Biomedical Engineering, 2008, 27(6):561-564.

[6] 王晶晶, 孙涛, 赵枫朝,等. 支持向量机在肺结节CT图像中的应用[J]. 北京生物医学工程, 2013, 32(5):528-530.

Wang JJ, Sun T, Zhao FC, et al. Application of support vector machine for pulmonary nodules in CT image[J]. Beijing Biomedical Engineering, 2013, 32(5):528-530.

[7] Sobieranski AC, Linhares RTF, Comunello E, et al. A fast gabor filter approach for multi-channel texture feature discrimination[M]//Martin CS，Kim SW. Progress in pattern recognition, image analysis, computer vision, and applications. London：Springer International Publishing, 2014:135-142.

[8] Daugman JG. Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters[J]. Journal of the Optical Society of America A Optics & Image Science, 1985, 2(7):1160-9.

[9] Jones JP, Palmer LA. An evaluation of the two-dimensional Gabor filter model of simple receptive fields in cat striate cortex[J]. Journal of Neurophysiology, 1987, 58(6):1233-1258.

[10] Liu SL, Niu ZD, Sun G, et al. Gabor filter-based edge detection: A note[J]. Optik - International Journal for Light and Electron Optics, 2014, 125(15):4120-4123.

[11] Zhu Y, Tan Y, Hua Y, et al. Feature selection and performance evaluation of support vector machine (SVM)-based classifier for differentiating benign and malignant pulmonary nodules by computed tomography[J]. Journal of Digital Imaging, 2010, 23(1):51-65.

[12] 胡强, 郝晓燕, 雷蕾. 基于遗传算法和BP神经网络的孤立性肺结节分类算法[J]. 计算机科学, 2016, 43(6A)：37-40.

Hu Q, Hao XY, Lei L. Solitary pulmonary nodules classification based on genetic algorithm and back propagation neural networks[J]. Computer Science, 2016, 43(6A) ：37-40.

[13] 纪国华，赵涓涓，潘玲.基于改进自生成神经网络的孤立性肺结节分类[J]. 太原理工大学学报，2015，（6）：754-759.

Guohua JI, Zhao J, Pan L. Solitary pulmonary nodule classification based on improved self-generating neural networks[J]. Journal of Taiyuan University of Technology, 2015，（6）：754-759.

[14] Mleczko WK, Kapu?ciński T, Nowicki RK. Rough deep belief network - application to incomplete handwritten digits pattern classification[M]// Information and Software Technologies.London： Springer International Publishing, 2015:400-411.

[15] Han F, Zhang G, Wang H, et al. A texture feature analysis for diagnosis of pulmonary nodules using LIDC-IDRI database[C]// 2013 IEEE International Conference on Medical Imaging Physics and Engineering. Shengyang：IEEE, 2014:14-18.

[16] 巩萍, 王姗姗, 罗举建. 基于稀疏自编码神经网络的肺结节特征提取及良恶性分类[J]. 医疗卫生装备, 2015, 36(12):7-10.

Gong P, Wang SS, Luo JJ. Feature extraction and benign or malignant classification of lung nodules based on sparse auto-encoder neural networks[J]. Chinese Medical Equipment Journal, 2015, 36(12):7-10.

[17] Raicu DS, Varutbangkul E, Cisneros JG, et al. Semantics and image content integration for pulmonary nodule interpretation in thoracic computed tomography[J]. Proceedings of SPIE - The International Society for Optical Engineering, 2007, 6512:65120S-65120S-12.

[18] Li W, Mao KZ, Zhang H, et al. Designing compact Gabor filter banks for efficient texture feature extraction[C]// International Conference on Control Automation Robotics & Vision. IEEE, 2011:1193-1197.

[19] Bianconi F, Fernández A. Evaluation of the effects of Gabor filter parameters on texture classification[J]. Pattern Recognition, 2007, 40(12):3325-3335.

[20] 王毅翔，邓敏. 孤立性肺结节的循证医学影像处理[J]. 临床与病理杂志, 2015, 35(2):169-174.

[21] 张春霞, 姬楠楠, 王冠伟. 受限波尔兹曼机[J]. 工程数学学报, 2015(2):159-173.

Zhang CX, Ji NN, Wang GW. Restricted boltzmann machines[J]. Chinese Journal of Engineering Mathematics, 2015(2):159-173.

[22] 周树森. 基于深度置信网络的分类方法[D]. 哈尔滨：哈尔滨工业大学, 2012.

Zhou SS. Deep belief networks based classification methods[D]. Harbin ：Harbin Institute of Technology, 2012.

[23] 聂生东, 邱建峰, 郑建立,等. 医学图像处理[M].上海：复旦大学出版社, 2010:192-193.

Nie SD, Qiu JF, Zheng JL, et al. Medical image processing[M], Shanghai：Fudan University Press, 2010:192-193.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录