Abstract

This paper describes a novel automatic skull-stripping method for premature infant data. A skull-stripping approach involves the removal of non-brain tissue from medical brain images. The new method reduces the image artefacts, generates binary masks and multiple thresholds, and extracts the region of interest. To define the outer boundary of the brain tissue, a binary mask is generated using morphological operators, followed by region growing and edge detection. For a better accuracy, a threshold for each slice in the volume is calculated using k-means clustering. The segmentation of the brain tissue is achieved by applying a region growing and finalized with a local edge refinement. This technique has been tested and compared to manually segmented data and to four well-established state of the art brain extraction methods.