High grade glioma segmentation in magnetic resonance imaging

Resumen

Through this work we propose a computational techniquefor the segmentation of magnetic resonance images(MRI) of a brain tumor, identified as high grade glioma(HGG), specifically grade III anaplastic astrocytoma. Thistechnique consists of 3 stages developed in the threedimensionaldomain. They are: pre-processing, segmentationand post-processing. The pre-processing stage usesa thresholding technique, morphological erosion filter(MEF), in gray scale, followed by a median filter and agradient magnitude algorithm. On the other hand, in orderto obtain a HGG preliminary segmentation, during thesegmentation stage a clustering algorithm called regiongrowing (RG) is implemented and it is applied to the preprocessedimages. The RG requires, for its initialization, aseed voxel whose coordinates are obtained, automatically,through the training and validation of an intelligent operatorbased on support vector machines (SVM). Due tothe high sensitivity of the RG to the location of the seed,the SVM is implemented as a highly selective binary classifier.During the post-processing stage, a morphologicaldilation filter is applied to preliminary segmentation generatedby RG. The percent relative error (PrE) is consideredby comparing the segmentations of the HGG, generatedmanually by a neuro-oncologist, with the dilated segmentationsof the HGG, obtained automatically. The combinationof parameters linked to the lowest PrE, allows establishingthe optimal parameters of each computationalalgorithms that make up the proposed computationaltechnique. The obtained results allow reporting a PrE of11.10%, which indicates a good correlation between themanual segmentations and those produced by the computationaltechnique developed.