基于SOM-PNN分类器的体数据概率分类及绘制

A SOM-PNN CLASSIFIER FOR PROBABILISTIC SEGMENTATION AND VISUALIZATION OF VOLUME DATA

概率分类是三维医学体数据绘制必不可少的预处理环节．本文提出的SOM－PNN分类器，以贝叶斯置信度为基础，给出概率分类结果，并用于三维体绘制，得到了良好的图像质量和较高的分类效率．传统的参数模型方法的主要缺点是预先假定的概率分布函数形式不一定符合待分类的数据．非参数模型方法，如PNN分类器，可以有效地克服参数模型的缺点，但其巨大的内存开销与低的分类速度使得用PNN作图像分类几乎不可行．SOM具有良好的自组织聚类能力，但无法直接给出概率分类结果．本文提出的SOM－PNN分类器在SOM聚类的基础上，利用PNN进行概率分类，结合了SOM自组织聚类和PNN概率分类的优势，同时克服了传统参数模型分类的局限性．实验结果证实了SOM-PNN分类器具有分类精度高、速度快及揭示细节的能力．

This paper presents a new probabilistic classifier, called SOM-PNN classifier, for volume data classification and visualization. The new classifier produces probabilistic classification withBayes confidence measure which is highly desirable in volume rendering. In the traditional parametric methods, the predefined probability function may be incapable of representing the data well-Non-parametric methods - such as PNN, can overcome the difficulty of the parametric methods.However, the huge storage and the slow evaluation time of the PNN algorithm make it impracticalin image processing with large training sets. SOM is efficient in clustering but it can not produceprobabilistic classification directly. Based on the trained SOM map, the SOM-PNN classifier performs the probabilistic classification using PNN algorithm. This combined use of SOM and PNNspeeds up the PNN evaluation significantly and increases its accuracy as well. The proposed SOM-PNN classifier has been used to segment the CT sloth data resulting in much better image qualitythan the parametric methods. Compared to the non-probabilistic classification by the SOM classifier, the probabilistically classified volume data results in more informative 3D rendering with moredetails. These demonstrate that the SOM-PNN classifier is a fast, accurate and probabilistic classifier for volume rendering.