Phantom based segmentation assessment for MRI images

This paper describes new effective methods for assessing the reliability of segmentation algorithms applied to encephalon extraction. The methods are based on the use of a brain phantom which provides a valuable global modeling of cerebral structures. Thanks to this realistic model, we create two original discrepancy measurements which are both applied to find the ideal parameters range for the segmentation algorithm proposed and to compare the accuracy of these methods. The most obvious application of this methodology lies in the extension of the results performed on the phantom to real images.