Classification of Brain Matters in MRI by Kernel Independent Component Analysis

An automatic segmentation system for MR imaging is necessary for studies and 3-dimensional visualization of anatomical structures in many clinical and research applications. Since conventional classification systems use a simple linear classifier, non-linear model is not taken into consideration. In this paper, we propose a new method based on kernel independent component analysis (KICA) for classification of phantom and clinical MR datasets. First, we extract kernel independent components from MR datasets by using KICA, and then the extracted components are used for classification of brain tissues. Since KICA, as a non-linear approach, can perform significant enhancement of brain MR datasets, the KICA-based classification method effectively classifies brain tissues and is computationally better than the conventional methods. The proposed method has been successfully applied to MR datasets and the classification performance has also been compared with conventional multi-spectral methods.