Enhanced Brain Tumor Classification from MRI Images Using MobileNetV2: A Transfer Learning Approach

Abstract

Brain tumors require being identified early and accurately, for prompt treatment and better patient outcomes. Conventional convolutional neural networks (CNNs) have shown encouraging results in the classification of tumors using magnetic resonance imaging (MRI); nevertheless, when trained on limited datasets of medical images, they frequently experience overfitting and significant computing costs. In this work, we introduce an efficient deep learning framework for classifying multiple types of brain tumors, leveraging MobileNetV2—a compact convolutional neural network architecture utilizing depthwise separable convolutions. A publicly accessible dataset of 3,264 MRI scans categorized  into four classes—glioma, meningioma, pituitary tumor, and no tumor—was used to train and assess the model. To improve model generalization, extensive preprocessing and data augmentation techniques were used. The proposed MobileNetV2-based model achieved higher reported performance than the evaluated architectures, including VGG16, ResNet50, InceptionV3, and the traditional CNN suggested in previous work, achieving 98.25% accuracy, 98.39% recall, and 98.69% AUC. Based on the results, MobileNetV2 is a promising option for implementation in real-time and resource constrained clinical contexts since it provides a highly accurate and computationally economical solution for brain tumor diagnosis.