Automatic MRI Bone Segmentation

Automatically segmenting bone tissue in MRI scans requires robustness against poor signal-to-noise ratios, highly inconsistent lighting conditions, and variability within bone tissues. Because of these difficulties, very little literature exists on automatic MRI bone segmentation. Current state-of-the-art methods are either semi-automatic or rely on databases of prior manual segmentations. Here, we propose a method that segments the radius and ulna bones using multiple passes of the Maximally Stable Extremal Regions (MSER) [1] algorithm on 2D slices of the volumetric MRI scans. On two MRI scans, our method achieved a Dice similarity coefficient (DSC) of 0.98, beating present stateof-the-art techniques.