This project compares classification models to discriminate tumor-damaged brain structures from normal ones using MRI images (Fig. 2A). Two types of models, including classical ML models (LDA, SVM, Random Forest, Gaussian Process) and deep learning (DL) models (MLP, ResNet, Vision Transformer) are tested. Classical ML models use compact input features derived from PCA and statistical properties whereas DL models use whole images per se. Vision Transformer model achieves the highest performance of 99.53% accuracy whereas the second best SVM achieves very close performance of 99.22%. Both models miss only 1~2% of tumor images. The result suggests that classical ML models can be effective when the appropriate features of brain images are extracted.