Classification of Brain Magnetic Resonance Images Based on Statistical Texture

Magnetic Resonance Image (MRI) is a medical technique commonly used by radiologists to visualize organ structures in humans without surgery, MRI provides a wealth of information about human soft tissue, which helps to diagnose brain tumors. Classifying normal and abnormal MRI images helps in early detection of brain tumors. Usually classifies MRI images using T2-weighted MRI images. However, to achieve perfect accuracy, sensitivity and specificity is very difficult. Fluid-attenuated inversion recovery (FLAIR) image is superior to T2-weighted image for detecting MS brain lesions. This study aims to classify FLAIR MRI images of the brain by classifying abnormal and normal MRI images based on statistical texture analysis, 44 abnormal datasets of BRATS 2017 training data and 4 normal datasets from Patient Contributed Image Repository (PCIR). At the beginning of the step, pre-processing of the image is followed by the histogram equalization method to extract the feature with statistical texture analysis by calculating the mean, variance, deviation, skewness, kurtosis, energy, entropy and smoothness. Finally, a multinomial logistic regression model with a ridge estimator is used to classify abnormal and normal MRI images and evaluated by k-fold validation. The results obtained from the proposed method of accuracy, sensitivity, and specificity reached 100%. This shows the method used to do a good classification in this case.