Brain image segmentation

Analysis of Segmentation Methods


2006-2009. Univ. of Bergen, Norway (Arvid Lundervold)


The objective of this project was to test new methods and compare state-of-the-art software for brain tissue segmentation and brain morphometry from 3D MR images. The methods we included in our study concerned local algorithms (LIMOS and Bergen), and segmentation algorithms provided by SPM5, FreeSurfer, and FSL. This effort was related to a multidisciplinary project on cognitive aging where the group in Bergen has acquired dual volume T1-weighted 3D brain images from 110 elderly subjects (age : 46-79 years) where also genotyping (APOE, Chrna4, BDNF) and results from extensive neuropsychological phenotyping is obtained. The best performing methods from the brain segmentation and morphometric studies were then be applied to this large image collection and numerical measures incorporated into statistical analysis for better insight in the aging process.

More specifically, the present project addressed the following issues :

  • Methods for performance evaluation
  • Optimal use of dual MR acquisitions (separate volume classification ? image registration ? value of single versus dual acquisition ?)
  • Influence of spatial inhomogeneities and noise in the MR signal, how to correct it ? Derive parameters (regional tissue volumes, cortical thickness, etc) that can be used in a broader statistical analysis with other biological and behavioral data
  • Make results (test-data, programs, evaluations) available on Web

Comparison of several brain segmentation methods