Fusing adaptive atlas and informative features for robust 3D brain image segmentation

• Published in: 2010 IEEE International Symposium on Biomedical Imaging: From Nano to Macro pp. 848 - 851
• Main Authors: Cheng-Yi Liu, Iglesias, Juan Eugenio, Toga, Arthur, Zhuowen Tu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.04.2010
• Subjects: Anatomical structure; Biomedical imaging; Biomedical informatics; Brain; discriminative; generative; Image segmentation; Magnetic resonance imaging; MRI; Neuroimaging; Robustness; segmentation; Shape; System testing
• ISBN: 9781424441259; 1424441250
• ISSN: 1945-7928
• DOI: 10.1109/ISBI.2010.5490119

It is an important task to automatically segment brain anatomical structures from 3D MRI images. One major challenge in this problem is to learn/design effective models, for both intensity appearances and shapes, accounting for the large image variation due to the acquisition processes by different machines, at different parameters, and for different subjects. Generative models study the explicit parameters for the generation process, and thus are robust against the global intensity changes; discriminative models are able to combine many of the local statistics, which are insensitive to complex and inhomogeneous texture patterns. In this paper, we propose a robust brain image segmentation algorithm by fusing an adaptive atlas (generative) and informative features (discriminative). We tested our algorithm on several datasets and obtained improved results over state-of-the-art systems.