<![CDATA[Accurate brain MRI classification is critical for early tumor diagnosis and computer-aided clinical decision support. Conventional convolutional neural networks (CNNs) are effective in learning deep hierarchical features but often struggle with intensity heterogeneity and partial volume effects inherent to MRI data. To address these limitations, this study proposes a hybrid Fuzzy C-Means–CNN (FCM–CNN) framework that integrates unsupervised soft clustering with deep feature learning. The fuzzy segmentation stage preserves boundary uncertainty by generating multi-channel membership maps, which are then fed into a CNN for robust classification. Evaluations conducted on the Kaggle brain MRI dataset (3,264 slices across four diagnostic categories) under Stratified 5-Fold Cross-Validation show consistent improvements over baseline models. The proposed FCM–CNN achieves a mean accuracy of 96.26% and Macro-F1 of 0.9622, surpassing both CNN-only and K-Means+CNN by +4.84% and +2.74% respectively. Ablation analysis confirms that soft memberships enhance discrimination between visually similar tumors, while statistical testing verifies that the gains are systematic and reproducible. Furthermore, the fuzzy membership maps provide interpretable visual cues, aligning with recent trends in explainable AI (XAI) for medical imaging. Overall, the FCM–CNN framework demonstrates that combining fuzzy logic with deep learning yields a balanced trade-off between performance, interpretability, and computational efficiency, making it promising for clinical-grade brain MRI analysis.]]>
Copyrights © 2025
