Classification of MRI Brain Tumor Images Based on Machine Learning Algorithm

Abstract

Brain Tumors (BTs) are serious medical conditions characterized by abnormal cellular growth in the brain. Magnetic Resonance Imaging (MRI) may be difficult and time-consuming for brain tumor identification and separation. This research automates the brain tumor classification, focusing on 4 diseases: gliomas, meningioma, and pituitary adenomas. Before employing Discrete Wavelet Transform (DWT) for tumor segmentation, MRI images should be converted to grayscale to improve speed and reduce computer complexity. PCA and Gray-Level Co-occurrence Matrix features are used to maximize feature extraction. The obtained features are classified using supervised K-Nearest Neighbours (KNN). The approach was tested using 2870 images, categorized as: 826 glioma, 822 meningioma, 827 pituitary adenoma, and 395 cancer-free images. The recommended approach used PCA and KNN to achieve 83% accuracy, however DWT and PCA together yielded 85% accuracy. The results show that the automated technique can quickly and accurately diagnose brain tumors.