摘要
提出了胰腺内镜超声图像的纹理特征提取与分类方法,可应用于胰腺癌内镜超声图像的计算机辅助诊断。对胰腺内镜超声图像采用数字图像处理算法提取9大类共69个纹理特征。使用类间距作为可分性判据,实现特征的初步筛选,之后使用顺序前进搜索算法进一步筛选特征,并由支撑向量机实现分类。对216例病例随机选取训练集和测试集,通过多次随机实验表明,本文提出的算法实现了较高的分类准确率,为胰腺癌的临床诊断提供有价值的参考意见。
The method of texture feature extraction and classification of pancreatic endoscopic uhrasonogra- phy（EUS） images is proposed, which can be applied to the computer-aided diagnosis of pancreatic EUS images. 69 texture features of 9 feature sets are extracted from the pancreatic EUS images according to the digital image pro- cessing algorithm. Dis ＂tance between classes is chosen as the separability criterion to select the features in the first step, then the Sequential Forward Selection algorithm is performed to further select the features. The classification is realized based on the support vector machine. The train set and test set are randomly chosen from 216 cases. Tests performed show that the proposed method result in a high classification accuracy, which will provide the phy- sician a valuable opinion on the diagnosis of pancreatic cancer.
出处
《生物医学工程学进展》
CAS
2008年第3期141-145,共5页
Progress in Biomedical Engineering
基金
上海市重点学科建设项目(B112)
关键词
胰腺癌
内镜超声
纹理特征
模式分类
Pancreatic Cancer
EUS
texture feature
pattern classification
