Enhancing Brain MRI Classification with Gray Wolf Optimization and Swin Transformers

With the rapid advancement of computational power, deep learning has become integral to medical imaging. Convolutional Neural Networks (CNNs) are widely applied to brain MRI scans for tumor classification due to their strong localization and recognition capabilities. However, their limited ability to capture long-range dependencies often leads to information loss and reduced performance in complex feature extraction. To overcome this, the Swin Transformer introduces a hierarchical attention mechanism with shifted windows that preserves spatial relationships and enhances the modeling of global contexts. This architecture improves the detection of intricate tissue textures and supports high-resolution reconstruction of MRI images. Furthermore, integrating the Swin Transformer with the Gray Wolf Optimization (GWO) algorithm enables fine-tuning of model parameters, enhancing accuracy and overall stability. The proposed hybrid framework demonstrates superior results in classification metrics, including accuracy, F1 score, precision, sensitivity, specificity, AUC, and minimized false rates. This synergy between Swin Transformer and GWO contributes to more reliable and efficient tumor classification, assisting radiologists in achieving faster and more precise diagnostic outcomes. Future work will focus on extending this model’s applicability and refining its adaptability for real clinical environments.