概率神经网络与差异演化在胶囊内窥图像出血识别中的应用  

Application of probabilistic neural network and differential evolution to bleeding detection in wireless capsule endoscopy images

作　　者：潘国兵[1] 颜国正[1] 张明卿[1] 邱祥玲[1] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《光学精密工程》2010年第6期1429-1435,共7页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.30570485);国家863高技术研究发展计划资助项目(No.2006AA04Z368)

摘　　要：为了解决在大量的胶囊内窥图像中寻找出血或相关病理特征这一难题,提出了一种智能自动识别胶囊内窥图像出血的方法。首先分析了胶囊内窥图像出血的颜色特征分布,然后利用差异演化算法(DE)对概率神经网络(PNN)进行了改进,使每个神经元传递函数具有不同的平滑参数。在此基础上提出了一种胶囊内窥图像出血智能识别的方法,并通过软件编程实现了该方法。实验结果表明,该软件能正确地识别出内窥图像中的出血区域并清晰地标示,用该方法测得的出血检测灵敏度和特异度分别为94%和87%,节省了图像识别时间,基本实现了胶囊内窥图像出血智能识别,可代替临床医生应用于胶囊内窥图像的初步检测。An automatic and intelligent computer aided bleeding detection technique is presented to recognize the bleeding regions and other pathological features in large amounts of images generated from a Wireless Capsule Endoscope（WCE）.Color features of the bleeding region in WCE images is extracted,and then the Probabilistic Neural Network （PNN） is improved by using differential evolution （DE） algorithm to offer the different smoothing parameters for each transfer function of neurons.Based on the improved PNN,the intelligent recognizing method is proposed and implemented through programming.The experimental results show that the bleeding regions in WCE images can be recognized correctly and marked clearly,and the sensitivity and the specificity of the method are measured as 94% and 87%,respectively.The intelligent bleeding detection method reduces the time-consuming for the WCE video detection and can help the clinician examine the gastrointestinal disease.

关 键 词：胶囊内窥镜 出血检测 概率神经网络 差异演化算法 

分 类 号：R445[医药卫生—影像医学与核医学] TP391.4[医药卫生—诊断学]