一种基于LBP特征提取和稀疏表示的肝病识别算法
发布时间：2018-08-08 21:32
【摘要】：由于肝脏超声图像具有回声不均匀、边缘模糊等缺点,肝脏疾病的无创诊断易受影响,而且目前临床基于肝脏超声图像的肝病诊断主要依靠医生的主观判断,其缺点为依赖医生主观经验且耗时,因此提出一种基于局部二值模式(LBP)特征提取和稀疏表示的肝病识别算法。从肝脏超声图像中提取感兴趣区域,使用LBP特征提取方法对感兴趣区域提取图像特征,将得到的特征进行字典训练,得到稀疏矩阵,最终采取支持向量机对其进行分类。实验样本均取自青岛大学附属医院肝胆科。实验1使用该方法对100个正常肝脏样本和100个肝硬化样本进行分类,准确率达到99.50%,实验2使用该方法对肝硬化、脂肪肝、肝血管瘤和肝癌4类样本共200个进行分类,AUC值分别为67.2%、65.1%、55.0%和62.6%。ROC曲线表明,提出的分类方法在准确率和泛化能力上均优于传统方法,有助于肝病的临床诊断。
[Abstract]:Due to the shortcomings of non-invasive diagnosis of liver diseases, such as uneven echo and blurred edges, the diagnosis of liver diseases based on liver ultrasound images mainly depends on the subjective judgment of doctors. The shortcomings of the algorithm are that it depends on the subjective experience of doctors and is time-consuming. Therefore, a liver disease recognition algorithm based on local binary pattern (LBP) feature extraction and sparse representation is proposed. The region of interest is extracted from the liver ultrasound image, and the region of interest is extracted by using the LBP feature extraction method. The obtained features are trained in dictionary to obtain sparse matrix, and finally the support vector machine is used to classify the region of interest. All samples were taken from Department of Hepatobiliary, affiliated Hospital of Qingdao University. Experiment 1 used this method to classify 100 normal liver samples and 100 liver cirrhosis samples, and the accuracy was 99.500.Experiment 2 used this method to classify liver cirrhosis and fatty liver. A total of 200 samples of liver hemangioma and liver cancer were classified with AUC of 67.2% and 65.1%, respectively, and the 62.6%.ROC curve showed that the proposed classification method was superior to the traditional method in accuracy and generalization ability, and was helpful for the clinical diagnosis of liver diseases.
【作者单位】： 青岛大学计算机科学技术学院;山东省数字医学与计算机辅助手术重点实验室;加州大学洛杉矶分校;
【基金】：国家自然科学基金(61303079;61305045)
【分类号】：R575;TP391.41
本文编号：2173095
[Abstract]:Due to the shortcomings of non-invasive diagnosis of liver diseases, such as uneven echo and blurred edges, the diagnosis of liver diseases based on liver ultrasound images mainly depends on the subjective judgment of doctors. The shortcomings of the algorithm are that it depends on the subjective experience of doctors and is time-consuming. Therefore, a liver disease recognition algorithm based on local binary pattern (LBP) feature extraction and sparse representation is proposed. The region of interest is extracted from the liver ultrasound image, and the region of interest is extracted by using the LBP feature extraction method. The obtained features are trained in dictionary to obtain sparse matrix, and finally the support vector machine is used to classify the region of interest. All samples were taken from Department of Hepatobiliary, affiliated Hospital of Qingdao University. Experiment 1 used this method to classify 100 normal liver samples and 100 liver cirrhosis samples, and the accuracy was 99.500.Experiment 2 used this method to classify liver cirrhosis and fatty liver. A total of 200 samples of liver hemangioma and liver cancer were classified with AUC of 67.2% and 65.1%, respectively, and the 62.6%.ROC curve showed that the proposed classification method was superior to the traditional method in accuracy and generalization ability, and was helpful for the clinical diagnosis of liver diseases.
【作者单位】： 青岛大学计算机科学技术学院;山东省数字医学与计算机辅助手术重点实验室;加州大学洛杉矶分校;
【基金】：国家自然科学基金(61303079;61305045)
【分类号】：R575;TP391.41
【相似文献】
相关期刊论文 前1条
1 刘金林;房颖;王涛;;肝硬化自发性细菌性腹膜炎患者血清LBP水平变化及意义[J];山东医药;2013年28期
相关硕士学位论文 前2条
1 宗韵;慢性胃炎相关几种典型病灶的特征提取与识别[D];浙江大学;2016年
2 庄展鸿;肝脏超声图像病灶区域分割算法的研究[D];北京化工大学;2016年
，本文编号：2173095
本文链接：https://www.wllwen.com/yixuelunwen/xiaohjib/2173095.html
最近更新
教材专著
