Brain Extraction Based on an Improved Charged Fluid Model in MRI

Abstract

Brain extraction is an important preprocessing step in many research and clinical applications. It aims to remove non-brain tissues (e.g., skull, scalp, and dura) and retain brain parenchyema in MR images. Among a number of brain extraction methods, deformable models are gaining attention as they provide robust abilities to deform contours under the guidance of geometric properties and image information to fit salient edges in images. In particular, the charged fluid model has been shown less critical than many existing deformable models in segmenting objects with sharp corners and cusps. This paper proposes a new brain extraction algorithm based on the charged fluid model. To improve the segmentation accuracy, a new image stopping force of using the local intensity difference along the normal line of the evolving curve is introduced. For brain extraction, this new stopping force is then incorporated into the governing equation weighted by a global intensity difference between the interior and exterior of the evolving contour. We have adopted the BrainWeb image data with various levels of noise and intensity non-uniformity to evaluate this new algorithm. Preliminary results indicated that our method successfully stopped the leakage comparing to the traditional charged fluid model and produced high segmentation accuracy across a number of brain MR images.KeywordsMRIsegmentationcharged fluid modelbrain extractiondeformable model