Comparison of CNN Transfer Learning Models for Brain Tumor Detection Based on MRI Images

Abstract

Brain tumors require early and accurate detection to support effective clinical decision-making. This study compares the performance of four transfer learning-based Convolutional Neural Network (CNN) models, namely DenseNet121, InceptionV3, MobileNet, and Xception, for brain tumor detection using MRI images. The dataset was preprocessed through resizing, normalization, and data augmentation, and all models were trained for 20 epochs using ImageNet pre-trained weights. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The experimental results show that all models achieved accuracies above 90%, with MobileNet outperforming the others by achieving an accuracy of 94.74% and precision, recall, and F1-score values of 0.95, 0.95 and 0,94. These findings indicate that lightweight CNN architectures can deliver superior performance for MRI-based brain tumor classification.