Accurate classification of brain tumors from MRI scans is crucial for effective diagnosis and treatment planning in neuro-oncology. This study presents a comprehensive framework leveraging advanced deep-learning techniques and visualization methods to achieve precise tumor classification. Our methodology involves merging two diverse datasets, performing data augmentation, and standardizing image dimensions to facilitate robust model training. We propose an attention-based multiscale fusion model, which integrates spatial attention and multi-scale fusion layers, achieving an accuracy of 99.17%, surpassing other models such as DenseNet201 (87.21%), InceptionV3 (82.30%), and MobileNet V3 (92.01%). Advanced visualization techniques, including Grad-CAM, enhance interpretability and confidence in diagnostic assessments. Moving forward, future work could explore integrating multi-modal imaging data and addressing challenges related to data scarcity and class imbalances to enhance diagnostic accuracy further and personalize treatment recommendations. Collaboration between computer scientists and clinicians is essential for seamless integration of AI systems into clinical workflows, ensuring reliable and safe deployment in real-world settings.