肝脏磁共振图像的混合聚类分割

Hybrid Clustering Method for Segmentation of Liver Magnetic Resonance Image

针对医学图像的自动分割,提出一种混合聚类方法。在对图像预处理后,将每个像素的邻域特征向量送入自组织特征映射网络(SOM)中进行训练。作为初步聚类的结果,SOM的输出典型向量根据命中图过滤,由层次合并聚类方法进一步处理。采用图像分割量化指数来确定聚类的最佳类别数;通过后处理得到最后分割结果,分析表明该方法是有效的。

A hybrid clustering method is proposed for automatic medical image segmentation.After pre-processing,local features of image pixels are extracted and sent to a self-organizing map（SOM）.The output prototypes of SOM are then filtered with the hits map.A hierarchical agglomerative clustering method is applied to the prototypes.The best segmentation is selected according to a quantitative image evaluation index.The segmented results are post-processed.The proposed method is effective and promising.