A pixel classification system for segmenting biomedical images using intensity neighborhoods and dimension reduction

We present an intensity neighborhood-based system for segmenting arbitrary biomedical image datasets using supervised learning. Because neighborhood methods are often associated with high-dimensional feature vectors, we explore a Principal Component Analysis (PCA) based method to reduce the dimensionality (and provide computational savings) of each neighborhood. Our results show that the system can accurately segment data in three applications: tissue segmentation from brain MR data, and histopathological images, and nuclei segmentation from fluorescence images. Our results also show that the dimension reduction method we described improves computational efficiency while maintaining similar accuracy.