Discriminative Confidence Estimation for Probabilistic Multi-atlas Label Fusion

Brain segmentation from magnetic resonance imaging (MRI) is an important preprocessing step for many neuroimaging studies, e.g., volumetry, cortical thickness, etc. For this task, automatic methods are desirable over manual segmentation since the latter is very time-consuming and subject to inter- and intra-rater variability. Although good outcomes can be achieved for the segmenation of the main tissues based only on image intensities  (Leemput et al., 1999; Ashburner and Friston, 2005; Shattuck et al., 2001), segmentation of anatomical structures (e.g., defined by their functional properties) renders intensity information insufficient and atlas priors become an imperative resource in order to accurately delineate such structu res.

Source: Medical Image Analysis - September 1, 2017 Category: Radiology Authors: Oualid M. Benkarim, Gemma Piella, Miguel Angel Gonz ález Ballester, Gerard Sanroma, for the Alzheimer’s Disease Neuroimaging Initiative Source Type: research