基于胰腺超声内镜图像多尺度分块的分类研究  

Classification based on multi-scaled partition in pancreatic endoscopic ultrasonography

作　　者：吴仪俊[1] 余建国[1] 刘岩[2] 金震东[2] 

机构地区：[1]复旦大学电子工程系,200433 [2]上海市长海医院消化科

出　　处：《上海医学影像》2011年第2期114-116,119,共4页Shanghai Medical Imaging

摘　　要：目的获取目标图像中癌症可能性较高区域的局部分布情况。方法综合胰腺超声内镜图像的多尺度分块方法以及现有的图像分类方法,根据多尺度分块的特点,对原用的分类方法进行改进,并提取23个图像纹理特征。结果将216幅胰腺超声内镜图像作为训练及测试样本进行多次随机实验,所得结果能显示图像中各个小块为癌症的可能性。结论此方法能为医生的临床诊断及放射粒子治疗的植入位置提供参考。Objective To obtain the distribution of regions with high probability of cancer.Methods The method of Multi-scaled partition and the original classification of pancreatic endoscopic ultrasonography（Eus）images was combined.The original method of classification was improved according to the new feature of partition,and 23 texture features were extracted.Results Totally 216 EUS images were used as training and testing samples and various random experiments were taken.Thus,the probability of cancer in each block was obtained and displaged.Conclusion This method can provide a reference position for implanting radioactive particles.

关 键 词：胰腺癌 内镜超声 纹理特征 多尺度分块 分类 

分 类 号：R735.35[医药卫生—肿瘤]