High grade glioma segmentation in magnetic resonance imaging

Through this work we propose a computational technique for the segmentation of magnetic resonance images (MRI) of a brain tumor, identified as high grade glioma (HGG), specifically grade III anaplastic astrocytoma. This technique consists of 3 stages developed in the threedimensional domain. They are: pre-processing, segmentation and post-processing. The pre-processing stage uses a thresholding technique, morphological erosion filter (MEF), in gray scale, followed by a median filter and a gradient magnitude algorithm. On the other hand, in order to obtain a HGG preliminary segmentation, during the segmentation stage a clustering algorithm called region growing (RG) is implemented and it is applied to the preprocessed images. The RG requires, for its initialization, a seed voxel whose coordinates are obtained, automatically, through the training and validation of an intelligent operator based on support vector machines (SVM). Due to the 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 morphological dilation filter is applied to preliminary segmentation generated by RG. The percent relative error (PrE) is considered by comparing the segmentations of the HGG, generated manually by a neuro-oncologist, with the dilated segmentations of the HGG, obtained automatically. The combination of parameters linked to the lowest PrE, allows establishing the optimal parameters of each computational algorithms that make up the proposed computational technique. The obtained results allow reporting a PrE of 11.10%, which indicates a good correlation between the manual segmentations and those produced by the computational technique developed.