TriCANet: Cross-Plane Attention and Swin Transformer Fusion for Multi-View Brain MRI Tumor Classification

Brain tumors are one of the most critical diseases when it comes to brain. Doctors usually rely on MRI scans to spot them early, but the process is not perfect. It takes time, experience and a lot of focus to go through each scan and sometimes mistakes can happen. Not to mention that every hospital doesn't have the experts needed on hand all the time. Recently, deep learning has been doing some impressive work in the medical field especially with brain tumor detection, most existing methods rely on single-plane MRI slices, lack contextual understanding across different anatomical views, and often struggle with generalization or interpretability. Here, we introduce a deep learning model that combines multiple techniques, Tri-Plane Contextual Attention Network (TriCANet), designed to leverage multi plane MRI data for robust tumor classification. As a part of preprocessing, we resized all MRI slices to $224 \times 224$ pixels, applied normalization and data augmentation techniques such as flipping, rotating and zooming are used to improve model robustness. TriCANet combines features extracted from axial, sagittal, and coronal views using a shared EfficientNetB0 backbone, ensuring consistent feature learning across slices. These features are passed through a novel Cross-Plane Attention Block to dynamically fuse spatially rich representations, which are further refined using a Swin Transformer Encoder to capture long-range dependencies. The final classification is done through a dense SoftMax layer, enabling the model to identify between glioma, meningioma, and pituitary tumors. We have used the Kaggle Brain Tumor MRI dataset contains 2D slices, our model adopts a 3D-inspired multi-planar architecture, simulating volumetric analysis. Achieving 95.3% accuracy, it outperforms models like VGG16, ResNet50, and InceptionV3, demonstrating strong performance and future readiness for real 3D MRI integration in clinical settings.