Multi-spectral brain tissue segmentation using automatically trained k-Nearest-Neighbor classification.

摘要: Conventional k-Nearest-Neighbor (kNN) classification, which has been successfully applied to classify brain tissue in MR data, requires training on manually labeled subjects. This manual labeling is a laborious and time-consuming procedure. In this work, new fully automated classification procedure presented, kNN automated. achieved by non-rigidly registering the data with probability atlas automatically select samples, followed post-processing step keep most reliable samples. The accuracy of method was compared rigid registration-based conventional kNN-based segmentation using subjects for segmenting gray matter (GM), white (WM) cerebrospinal fluid (CSF) 12 sets. Furthermore, all methods, performance assessed when varying free parameters. Finally, robustness evaluated 59 non-rigid registration significantly more accurate than registration. For both trained classifier, difference observers not larger inter-observer variability types. From study, it clear that, given an appropriate optimal parameters, our automated, gives robust results. A similarity index used comparison kNN. indices were 0.93, 0.92 0.92, CSF, GM WM, respectively. It can be concluded that may replace segmentation, thus without feasible.