Salient features useful for the accurate segmentation of masticatory muscles from minimum slices subsets of magnetic resonance images

Abstract

The masticatory muscles play a critical role in the mastication system and directly affect one’s ability to chew and smile. We describe a new approach for obtaining patient-specific human masticatory muscle surface renderings from magnetic resonance images (MRI) of the head. We determine the set of dominant slices, from training data, that together best represent the salient features of the three-dimensional muscle shape. Candidates for the dominant slices are identified by shape- and area-based criteria, and this is followed by fuzzy C-means clustering to determine the slices that are selected. Two-dimensional segmentation is carried out on these dominant slices on the test data, with shape-based interpolation then applied to construct accurate muscle surface renderings. Performance evaluation using a leave-one-out method results in average overlap indices of greater than 90%, indicating that there is consistency between the surface renderings and manual contour tracings provided by an expert radiologist.