Brain tumor diagnosis via magnetic resonance imaging (MRI) plays a vital role in clinical neuroscience but remains hindered by challenges such as poor generalization and limited model interpretability. To address these issues, this paper proposes a novel dual-task framework that performs both brain tumor segmentation and multiclass classification by integrating quantum-inspired channel and spatial attention modules within a U-Net backbone. These attention mechanisms, influenced by quantum principles such as phase encoding and interference, enable the model to extract expressive and task-specific features for enhanced tumor boundary delineation and accurate tumor type prediction. Preprocessing steps including resizing, CLAHE-based contrast enhancement, and intensity normalization were employed. The model was trained on the Figshare dataset, achieving a Dice score of 0.85 and IoU of 0.81 for segmentation, and 97% accuracy, 96% precision, and a ROC-AUC of 0.99 for classification. To support human–AI collaboration, explainability techniques such as Grad-CAM and LIME were applied, allowing clinicians to visualize prediction-relevant regions and validate model decisions. The results demonstrate that the proposed framework not only delivers accurate predictions but also fosters interpretability and trust, making it a viable AI assistant to radiologists in brain tumor analysis.
Hossain et al. (2026) studied this question.