Automatic segmentation of epidural hematomas using a computational technique based on intelligent operators: a clinical utility

Abstract

This paper proposes a non-linear computational techniquefor the segmentation of epidural hematomas (EDH), presentin 7 multilayer computed tomography brain imaging databases.This technique consists of 3 stages developed in thethree-dimensional domain, namely: pre-processing, segmentationand quantification of the volume occupied by each ofthe segmented EDHs. To make value judgments about theperformance of the proposed technique, the EDH dilated segmentations,obtained automatically, and the EDH segmentations,generated manually by a neurosurgeon, are comparedusing the Dice coefficient (Dc). The combination of parameterslinked to the highest Dc value, defines the optimal parametersof each of the computational algorithms that makeup the proposed nonlinear technique. The obtained resultsallow the reporting of a Dc superior to 0.90 which indicatesa good correlation between the manual segmentations andthose produced by the computational technique developed.Finally, as an immediate clinical application, considering the automaticsegmentations, the volume of each hematoma is calculatedconsidering both the voxel size of each database and thenumber of voxels that make up the segmented hematomas.