Feature extraction on brain magnetic resonance imaging

Abstract

Brain tumor is a defying death disease, causing millions of deaths each year around the globe. Its treatment is utterly dependent on early and accurate detection of the tumor and abnormal masses present in the brain, as it will increase patients' survival rate significantly and thus lower mortality rate. Manual diagnosis of brain tumors by radiologists and experts usually proves to be tedious, time consuming, prone to error and a very costly process, making early detection less occurrent. As a result, the introduction and the development of image processing, computer-based techniques and accurate models for the detection of brain tumors became a research field of a significant importance. MR Imaging is the most common technique used for brain tumor detection. Several techniques have been proposed throughout the years to automate this process.

in this study, three different approaches to extract features from the images have been used. The first approach was to extract statistical features directly from gray-level co-occurrence matrix (GLCM) of the whole and cropped images. In the second approach, we have used the Discrete Cosine Transform (DCT) to the whole and cropped images then extracting statistical features. In the third approach, we have used the Discrete Wavelet Transform (DWT). In the last approach, the combination of the Discrete Cosine Transform with DWT was used.