Performance analysis of automatic techniques for tissue classification in magnetic resonance images of the human brain

摘要: Classification of Magnetic Resonance (MR) images the human brain into anatomically meaningful tissue labels is an important processing step in many research and clinical studies neurology. The medical imaging community presented with a wide choice classification algorithms from artificial intelligence pattern recognition. This thesis describes development controlled test environment, where different were implemented their performance evaluated context. Furthermore, mechanism for automating supervised proposed through use priori knowledge neuro-anatomy, form probability maps. results obtained automated methods compared favorably to those supervision. five (Artificial Neural Networks, Bayesian, k-Nearest Neighbors, C4.5 decision tree, Minimum Distance) two unsupervised (Hard C Means, Fuzzy Means) under varying conditions MR artifacts. Artificial networks classifier was observed be best overall performer