An Optimized U-Net and MobileNetV2 Framework for Accurate and Efficient Brain Tumour Classification from MRI Images

Abstract

Accurate brain tumour identification from magnetic resonance imaging (MRI) requires models that can effectively localize tumour regions while maintaining high classification performance and computational efficiency. This study presents an integrated U-Net–MobileNetV2 framework for automated brain tumour analysis using MRI images. The proposed pipeline incorporates image thresholding, morphological processing, contour-based cropping, bicubic resizing, normalization, and data augmentation prior to model training. U-Net is employed to segment tumour regions and preserve diagnostically relevant spatial information, while MobileNetV2 performs lightweight classification using segmentation-guided representations. The dataset consists of 512 × 512-pixel brain MRI images divided into 80% training, 10% validation, and 10% testing subsets. Model performance is evaluated using accuracy, precision, recall, and F1-score and compared with CNN, Random Forest, Naïve Bayes, and Support Vector Machine classifiers. The proposed framework achieves 99.52% accuracy, 97.57% precision, 98.05% recall, and 97.81% F1-score, outperforming all evaluated baseline models. Compared with the strongest baseline, SVM, the proposed method improves accuracy by 1.43 percentage points. These results demonstrate that combining segmentation-driven tumour localization with lightweight deep feature learning provides an effective framework for accurate and computationally efficient MRI-based brain tumour classification.