BEaST: Brain extraction using multiresolution nonlocal segmentation.

Simon Eskildsen (1) , Pierrick Coupé (1) , Kelvin K. Leung (2) , Vladimir Fonov (1) , Nicolas Guizard (1) , Shafik N. Wassef (1) , Lasse Riis Ostergaard (3) , Louis D. Collins (1)

1 MNI - McConnell Brain Imaging Centre
2 UCL - University College of London [London]
3 Department of Health Science and Technology

Brain extraction is an important step in the analysis of brain images. Variability in brain morphology and intensity characteristics due to different imaging sequences makes the development of a general pur- pose brain extraction algorithm challenging. Purpose: To address this issue, we propose a new robust method (BEaST) for brain extraction. Methods: The method is based on nonlocal segmentation embedded in a multiresolution framework. A library of 50 priors are semi-automatically constructed from the NIHPD, ICBM, and ADNI databases. Results: A mean Dice coefficient of 0.9834±0.0053 is obtained when performing leave-one-out cross validation. Validation using the online available Seg- mentation Validation Engine resulted in a top ranking position with a mean Dice coefficient of 0.9781±0.0047. Conclusions: The segmentation accuracy of the method is comparable to that of a recent label fusion ap- proach, while being 40 times faster and requiring a much smaller library of priors.

Mots clés: nonlocal means; brain extraction; patch-based segmentation

Domaines: Analyse numérique [cs.NA]; Imagerie médicale; Traitement des images [eess.IV]; Traitement du signal et de l'image [eess.SP]; Neurosciences; Traitement du signal et de l'image [eess.SP]; Applications [stat.AP]

Simon Eskildsen, Pierrick Coupé, Kelvin K. Leung, Vladimir Fonov, Nicolas Guizard, et al.. BEaST: Brain extraction using multiresolution nonlocal segmentation.. MICCAI Workshop on Multi-Atlas Labeling and Statistical Fusion, Sep 2011, Toronto, Canada. pp.97-108. ⟨hal-00614307⟩