This study proposes a new method on “brain image segmentation”. Algorithm is based on the Multistable Cellular Neural Networks (Multistable-CNN). Evaluations are performed on well-known BRAINWEB database. The metrics for performance evaluations of the algorithm were JACCARD, DICE, TPF and TNF. The results are compared with the SPM8, FSL FAST and Brain Suite software packages. The Multistable CNN algorithm used in this study differs from conventional CNN algorithm. The Multistable CNN’s can perform multiple segmentations in a single run. Algorithm is simple and doesn’t require complex calculations. Evaluations show that proposed algorithm’s performance is adequate. Algorithm also does not require any atlas or image registration.
Prepared by Burak Yılmaz, Akif Durdu, Güzin Özmen |