Brain Tumor Detection and Classification Using MobileNetV2

Brain tumor classification using Magnetic Resonance Imaging (MRI) is a crucial task in medical diagnostics, directly influencing treatment plans and patient outcomes. Manual evaluation of MRI images is time-consuming, prone to inter-observer variability, and requires expert analysis. To address these limitations, this study proposes a deep learning model based on the MobileNetV2 architecture for binary classification of brain MRI scans into glioma vs. non-tumor categories. This data set comprises 3,621 T1-weighted contrast-enhanced MRI images; 2,916 were used for training and 705 for evaluation. To enhance the model's generalization, image preprocessing methods such as normalization, resizing, and data augmentation were implemented. The optimizer was Adam, and binary cross-entropy was used as the loss function. The model was evaluated using performance metrics, such as precision, recall, F1-score, accuracy, and a confusion matrix, and an ROC-AUC score of 0.99. The findings indicated that the training accuracy was 98.38. These findings support the high classification capacity of the model and its application in clinical decision-support systems (especially in environments where computational resources are limited). The novelty of this work is its use of a lightweight architecture that achieves high classification accuracy while maintaining computational efficiency.