Automatic segmentation of a cerebral glioblastoma using a smart computational technique

Abstract

We propose an intelligent computational technique for theimage segmentation of a type IV brain tumor, identified asmultiform glioblastoma (MGB), which is present in multi-layercomputed tomography images. This technique consists of3 stages developed in the three-dimensional domain. Theyare: pre-processing, segmentation and validation. During thevalidation stage, the Dice coefficient (Dc) is considered inorder to compare the segmentations of the MGB, obtainedautomatically, with the segmentations of the MGB generatedmanually, by a neuro-oncologist. The combination of parameterslinked to the highest Dc, allows to establish the optimalparameters of each of the computational algorithms thatmake up the proposed nonlinear technique. The obtained resultsallow to report a Dc higher than 0.88, validating a goodcorrelation between the manual segmentations and thoseproduced by the computational technique developed.