Application of Machine Learning Methods for Automatic Brain Tumor Recognition in MRI Images

Abstract

This paper explores machine learning methods applied for the automatic recognition of brain tumors in MRI 
images. An open BraTS 2021 dataset with expert annotations was used. Both classical and deep learning models were 
analyzed, including SVM, Random Forest, CNN, and U-Net. Evaluation was based on accuracy, recall, F1-score, Dice, 
and IoU. The ResNet50 model with transfer learning achieved the best classification accuracy (94%), while U-Net 
performed best in segmentation (Dice 0.87). Study limitations and future directions are discussed. The results may support 
clinical decision-making and improve diagnostic efficiency.