A new clustering algorithm for segmentation of magnetic resonance images

摘要: The major goal of this dissertation is to present a new clustering algorithm using information theoretic measures and apply the segment Magnetic Resonance (MR) Images. Since MR images are highly variable from subject subject, data driven segmentation methods seem appropriate. We developed evaluation function based on theory that outperforms previous algorithms, cost works as valley seeking algorithm. optimization difficult because its stepwise nature existence local minima, we an improvement K-change used commonly in problems. When applied nonlinearly separable data, performed with very good results, was able find nonlinear boundaries between clusters without supervision. The brain successful results. A feature set created entropy small blocks input image. Clustering whole image computationally intensive. Therefore, section first train Afterwards, rest clustered results obtained training by distance measure proposed. easy calculations simplified choosing proper which does not require numerical integration.