A PCA-based approach to the representation and recognition of MR brain midsagittal plane images

The midsagittal plane (MSP) of brain images is a key landmark used for many processing and analysis procedures associated with diagnosis, and intervention planning for the brain. In this paper, we propose to represent MSP images with features from principal component analysis (PCA). MSP images are extracted from 43 brain MRI volumes of different subjects. Images are represented as linear combinations of a set of eigen images which define the eigenspace that best describes the variation distribution of the set of images. The results of tests using planes extracted with slight orientational variations about the MSP show 100% recognition accuracy for planes varied under 0.5 degrees, and its robustness to MSP extraction error within variations of 1 degree in magnitude. The validation of this method with different MSP images yields a recognition accuracy of 88.6%.