Brain Tumor Classification Using Statistical and Texture Features: An Evaluation of Machine Learning Models

Abstract

Classification of brain tumors is still an essential problem in medical imaging, where the differentiation between tumor and non-tumor tissues may enhance the diagnosis and treatment processes. In this paper, the potential of different first-order statistical and texture-based features of tumors in the brain. The dataset comprises five first-order features (mean, variance, standard deviation, skewness, and kurtosis) and eight texture features (e.g., contrast, energy, and entropy) that capture essential characteristics in pixel intensity and spatial relationships within brain scans. These features are harnessed to enhance the differentiation between tumor and non-tumor classification tasks. In our analysis, we compared Decision Tree, Random Forest and Gaussian Naive Bayes. Our results indicated that RF and DT were the best models, with accuracies of about 0.975. The study's findings also show that statistical and texture features are essential in enhancing classification performance, hence the significance of highly correctly classified instances in detecting brain tumors.