Introduction: The brain tumor is an abnormal growth of tissue in the brain, which is one of the most important challenges in neurology. Brain tumors have different types. Some brain tumors are benign and some brain tumors are cancerous and malignant. Glioblastoma Multiforme (GBM) is the most common and deadliest malignant brain tumor in adults. The average survival rate for people with this type of brain tumor is about 15 months. Brain tumors are more common in men and are more dangerous. The most important diagnostic modality for tumor detection is magnetic resonance imaging (MRI). MRI is a non-invasive diagnostic method that provides anatomical images of the tumors. In recent decades, advanced MRI techniques have been increasingly developed to better tumor detection. One of these methods is Magnetic Resonance Spectroscopy (MRS) imaging. The MRS technique is used to study human brain metabolites and evaluate the neurochemical profile of the brain tissue. Unlike the MRI, MRS is able to grade the tumor. Depending on the grade and metabolites of the tumor, MRS can complement MRI images in medical diagnoses. The purpose of this study is to use machine learning to discriminate between normal and tumorous voxels in MRS data which can lead to a reduction in human error in the diagnosis of neurologist, radiologist, neuroscientists and etc.
Materials and Method: According to the neurologist's comment, magnetic resonance spectroscopy imaging was performed on 7 patients with GBM at the imaging center of Imam Khomeini Hospital in Tehran. The radiologist labeled all the tumor and normal voxels. Preprocessing step, including baseline correction and water suppression was performed by TARQUIN software. In processing step, signals of each voxel were extracted and the concentration of the metabolites was calculated. For classification of normal and tumorous voxels, Support Vector Machine was done using Statistics and Machine Learning Toolbox by MATLAB software.
Results: For classifying the data, the support vector machine (SVM) was used. The results of classifiers showed 87% accuracy, 82% specificity and 93% sensitivity. For classify these data, the Gaussian kernel was used. Using the linear kernel, the accuracy obtained 63%, specificity obtained 56% and sensitivity obtained 68%.
Conclusion: The results showed that, the accuracy for SVM with Gaussian kernel is significantly higher than SVM with linear kernel. This result shows that; this dataset may have nonlinear distribution. Therefore, the nonlinear classifiers may show better results than linear classifiers.