核最小模最小平方误差方法医学图像识别算法  

Kernel Minimum Mean Square Error Method for Medical Image Recognition

作　　者：夏开建 靳勇[2] 

机构地区：[1]苏州大学附属常熟医院(常熟市第一人民医院)信息科,江苏常熟215500 [2]常熟理工学院计算机工程与科学学院,江苏常熟215500

出　　处：《中国医疗设备》2018年第2期73-76,80,共5页China Medical Devices

基　　金：苏州市科技发展计划项目(SYSD2015014);常熟市科技局资助性项目(CS201503)

摘　　要：最小平方误差方法(Least Square Error,MSE)因其在对模式分类中所具备的有效性和高效性,在模式识别领域得到广泛的应用。同时基于核方法的非线性理论的不断成熟,针对医学图像识别通常存在的非线性可分问题,提供了一种有效的解决途径。本文将两者结合,并针对MSE中存在的投影向量"超定"的问题加以分析和改进,提出了这种基于核理论的最小模最小平方误差方法(Kernel Minimal Mean Square Error,KMNMSE),并建立了一种一般的MNMSE分类器模型。最后通过在CT医学图像上做了大量的实验,实验结果与其他方法的比较,验证了本文所提出方法的有效性。The least square error （MSE） method is widely used in pattern recognition because of its effectiveness and efficiency in pattern classification. At the same time, the nonlinear theory based on kernel trick （kernel） is mature, which provides an effective solution to the nonlinear and separable problems commonly existing in medical image recognition. In this paper, we put forward the method of kernel minimal mean square error （KMNMSE） by combining these two methods based on the core theory and solved the problems existed ＂in the presence of over-determined＂ during MSE method. A general MNMSE classification model was then established. Finally, a large number of experiments were done on CT medical images, and the experimental results were compared with other methods. Our results verify the effectiveness of the proposed method.

关 键 词：最小平方误差 核方法 核最小模最小平方误差方法 核主分量分析 医学目标识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]