Classification of brain stem glioma tumor grade based on MRI findings using support vector machine

Brain stem glioma is one of the brain tumors forming 10 to 20 percentages of tumors in children and 2 percentages of tumors in adults. It has two grades including high grade and low grade. Relatively, grade diagnosis is done by biopsy. The goal of this study is presenting a classification model based on MRI findings in order to diagnose glioma tumor and also investigating the effect of MRI findings on tumor’s grade.
In this cross-sectional study, we utilized MRI and pathological information of all 96 patients with glioma tumor in stereotactic biopsy ward of Shohadaye Tajrish hospital (Iran) between 2006-2012. For analysis of data, support vector machine as a precise classification model has fitted which is suitable for dataset with vast predictors or several class variables with low frequencies in some of them. This model has fitted in R software, 3.3.1 version.
The validation shows 93 percent total accuracy, 90 percent sensitivity and 93 percent specifity of support vector machine classifier model. Notably, the coefficients show positive correlation between headache, tumor spread in cord, homogeneous appearance, Cystlike appearance, ISO signal in T1 and T2 and low grade tumor and positive correlation between pons conflict, Tumor spread in thalamus, well defined appearance, necrosis appearance, hypersignal in T2 and heterogeneous enhancement with high grade tumor.
Support vector machine classification model based on MRI has high accuracy in tumor grade diagnosis.