Attention-Enhanced Swin Transformers for Robust Brain Tumor Classification Under Patient-Level Data Splitting
Abstract
Contemporary brain tumor classification systems report accuracies exceeding 98%, yet such metrics are artificially inflated by dataset partitioning flaws. Conventional image-level splitting allows identical patient scans in both training and testing sets, enabling networks to memorize patient-specific features rather than tumor patterns. We enforce patient-level separation where each patient remains in a single partition. Evaluation of five architectures reveals degradations reaching 3.71% under patient-level splitting, validating widespread data leakage. We propose an Attention-Enhanced Swin Transformer integrating hierarchical windowed attention with Convolutional Block Attention Modules. Our architecture achieves 96.82% accuracy with 1.23% degradation—the smallest gap among methods. Gradient-weighted activation mapping confirms attention on tumor regions rather than anatomical artifacts, establishing reliable feature extraction for trustworthy medical AI.
Key Methodologies & Contributions
- Exposing Data Leakage: We demonstrated that conventional 70/15/15 random partitioning creates cross-patient contamination, artificially boosting accuracy metrics in baseline models.
- Strict Patient-Level Partitioning: We utilized the Figshare Brain Tumor MRI dataset and organized scans strictly by patient identity, guaranteeing each patient’s data is isolated to a single partition for genuine clinical generalization.
- Attention-Enhanced Backbone: We integrated Convolutional Block Attention Modules (CBAM) into a Swin foundation backbone to prevent shortcut learning.
- Dual-Stage Feature Refinement: We utilized sequential channel and spatial attention to explicitly prioritize tumor-discriminative features while suppressing patient-specific anatomical signatures (e.g., skull boundaries and ventricular configurations).