Search

GBM is a markedly heterogeneous brain tumor consisting of three main volumetric phenotypes identifiable on magnetic resonance imaging: necrosis (vN), active tumor (vAT), and edema/invasion (vE). The goal of this study is to identify the three glioblastoma multiforme (GBM) phenotypes using a texture-based gray-level co-occurrence matrix (GLCM) approach and determine whether the texture features of phenotypes are related to patient survival. MR imaging data in 40 GBM patients were analyzed. Phenotypes vN, vAT, and vE were segmented in a preprocessing step using 3D Slicer for rigid registration by T1-weighted imaging and corresponding fluid attenuation inversion recovery images. The GBM phenotypes were segmented using 3D Slicer tools. Texture features were extracted from GLCM of GBM phenotypes. Thereafter, Kruskal-Wallis test was employed to select the significant features. Robust predictive GBM features were identified and underwent numerous classifier analyses to distinguish phenotypes. Kaplan-Meier analysis was also performed to determine the relationship, if any, between phenotype texture features and survival rate. The simulation results showed that the 22 texture features were significant with p value <0.05. GBM phenotype discrimination based on texture features showed the best accuracy, sensitivity, and specificity of 79.31, 91.67, and 98.75 %, respectively. Three texture features derived from active tumor parts: difference entropy, information measure of correlation, and inverse difference were statistically significant in the prediction of survival, with log-rank p values of 0.001, 0.001, and 0.008, respectively. Among 22 features examined, three texture features have the ability to predict overall survival for GBM patients demonstrating the utility of GLCM analyses in both the diagnosis and prognosis of this patient population.

To isolate the brain from non-brain tissues using a fully automatic method may be affected by the presence of radio frequency non-homogeneity of MR images (MRI), regional anatomy, MR sequences, and the subjects of the study. In order to automate the brain tumor (Glioblastoma) detection, we proposed a novel approach of skull stripping for axial slices derived from MRI. Then, the brain tumor was detected using multi-level threshold segmentation based on histogram analysis. Skull-stripping method, was applied by adaptive morphological operations approach. This is considered an empirical threshold by calculation of the area of brain tissue, iteratively. It was employed on the registration of non-contrast T1-weighted (T1-WI) and its corresponding fluid attenuated inversion recovery sequence. Then, we used multi-thresholding segmentation (MTS) method which is proposed by Otsu. We calculated the performance metrics based on the similarity coefficients for patients (n = 120) with tumor. The adaptive algorithm of skull stripping and MTS of segmented tumors were achieved efficient in preliminary results with 92 and 80 % of Dice similarity coefficient and 0.3 and 25.8 % of false negative rate, respectively. The adaptive skull stripping algorithm provides robust skull-stripping results, and the tumor area for medical diagnosis was determined by MTS., Competing Interests: The authors declare that they have no competing interests.