基于随机森林的脑磁共振图像分类

特征选择和分类算法是影响脑磁共振图像分类精度的2个最主要的因素。随机森林算法作为一种优秀组合分类器逐渐成为近年来研究的热点,通过加权脑磁共振3种（T1、T2、PD加权像）图像,采用非统一滑动窗口尺寸提取二维图像的纹理特征、形状特征、HAAR特征、灰度特征以及边缘检测算子、最大类间方差（OTSU）作为随机森林算法的输入特征,从而分类出图像的10类组织。经过对加拿大蒙特利尔神经科学研究院提供的脑仿真核磁共振图像实验,随机森林算法对二维脑MR图像的分类精度可以达到94%以上。

Magnetic resonance image classification of brain based on random forest

Feature selection and classification algorithm are two major factors that affect the classification accuracy of the magnetic resonance image of brain. The random forest algorithm as an excellent combination of classifier has became a research hotspot gradually in recent years. For the three magnetic resonance images of brain （ T1,T2, and PD-weighted image）, non-uniform sliding window size to extract the two-dimensional image texture features, shape features, HAAR features, gray features, edge detection operator, and the maximal variance between-class （OTSU） are used as the input characteristics of the random forest algorithm, and then the 10 organizations of the image are sorted finally. The random forest algorithm has a very high accuracy which is more than 94% for classification of two-dimensional brain MR image by the experiment of brain simulation on magnetic resonance image from Canada Neuroscience Research Institute.