An Algorithm of the LBP Feature Extraction Method Combining Sparse Representation in Liver Diseases Recognition
Han Xiuzhi1, Zhao Ximei1,2* , Yu Kexin3, Wang Guodong1
1College of Computer Science & Technology, Qingdao University, Qingdao 266071, Shandong, China 2Shandong Province Key Laboratory of Digital Medicine and Computer Aided Surgery, Qingdao 266000, Shandong, China 3University of California, Los Angeles, Los Angeles CA 90015, USA
Abstract：Because of the existence of the uneven echo, the dim edge and other factors under the ultrasonic environment, the diagnose accuracy of liver diseases can be influenced. Besides, the clinical diagnosis for liver diseases based on liver ultrasonic figure is mainly based on the conventional visual assessment that relies on the subjective experience of radiologists, leading to inaccurate and inadequate results. This study therefore proposed alocal binary pattern feature extraction method combining sparse representation algorithm. The proposed method extracted theregions of interest from liver ultrasonic figures, LBP method for feature extraction, dictionary learning method for training, andsupport vector machine for classification. In order to verify this approach, experiments were carried out by selecting samples from department of hepatobiliary surgery of affiliated hospital of Qingdao University. Results showed in the first experiment that the accuracy rate reached 99.50% of 100 normal liver samples and 100 liver cirrhosis samples. And results showed in the second experiment that the AUC of liver cirrhosis, fatty liver, hemangioma and liver cancer reached 67.2%, 65.1%, 55.0% and 62.6% separately of total 200 samples. The comparison between the proposed method and conventional methods, via receiver operating characteristic curve, demonstrated that the proposed method possessed the advantages both in accuracy and generalization performance. The proposed method would be helpful for clinical diagnosis of liver diseases.
韩秀芝, 赵希梅, 于可歆, 王国栋. 一种基于LBP特征提取和稀疏表示的肝病识别算法[J]. 中国生物医学工程学报, 2017, 36(6): 647-653.
Han Xiuzhi, Zhao Ximei, Yu Kexin, Wang Guodong. An Algorithm of the LBP Feature Extraction Method Combining Sparse Representation in Liver Diseases Recognition. Chinese Journal of Biomedical Engineering, 2017, 36(6): 647-653.
[1] 严春满. 图像稀疏编码算法及应用研究[D]. 西安:西安电子科技大学, 2012.
[2] Rodriguez F, Sapiro G. Sparse representation for image classification: learning discriminative and reconstructive nonparametric dictionaries[EB/OL]. http://hdl.handle.net/11299/179948,2008-06/2016-12-20.
[3] Wright J, Yang AY, Ganesh A, et al. Robust face recognition via sparse representation [J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2008, 31(2):210-227.
[4] Ogawa K, Fukushima M, Kubota K, et al. Computer-aided diagnostic system for diffuse liver diseases with ultrasonography by neural networks[J]. IEEE Transactions on Nuclear Science, 1998, 45(6):3069-3074.
[5] Horng MH, Sun YN, Lin XZ. Texture feature coding method for classification of liver sonography [J]. Computerized Medical Imaging & Graphics, 2002, 26(1):33-42.
[6] Yeh W C, Huang S W, Li P C. Liver fibrosis grade classification with B-mode ultrasound.[J]. Ultrasound in Medicine & Biology, 2003, 29(9):1229-1235.
[7] Sookoian S, Castaño G, Burgueño AL, et al. A diagnostic model to differentiate simple steatosis from nonalcoholic steatohepatitis based on the likelihood ratio form of Bayes theorem [J]. Clinical Biochemistry, 2009, 42(7):624-629.
[8] Wu CM, Chen YC, Hsieh KS. Texture features for classification of ultrasonic liver images [J]. IEEE Transactions on Medical Imaging, 1992, 11(2):141-152.
[9] Sookoian S, Castaño G, Burgueño AL, et al. A diagnostic model to differentiate simple steatosis from nonalcoholic steatohepatitis based on the likelihood ratio form of Bayes theorem [J]. Clinical Biochemistry, 2009, 42(7): 624-629.
[10] 赵志刚,吴鑫,洪丹枫,等. 基于信息熵的GLBP掌纹识别算法[J].计算机科学,2014,41(8): 293-296.
[11] Peter Harrington. 机器学习实战[M]. 北京:人民邮电出版社, 2013.89-91.
[12] 周国辉, 汪源源, 王威琪,等. 基于超声图像特征的肝硬化分析系统[J]. 声学技术, 2003, 22(4):288-291.
