EffiRes-BT: A Hybrid Architecture for Brain Tumor Classification

Brain tumors are among the most critical and medically challenging ailments because of the occasionally lethal consequences and complexity of treatment methods. Despite advancement in deep learning, existing models often struggle while diagnosing brain tumors leading to misclassification in cases with overlapping morphological features. This research proposes a new clinically reliable and high diagnostic accuracy hybrid architecture, EffiRes-BT which combines EfficientNet-B0 and Resnet-50 architectures to extract tumor characteristics. For evaluation, its effectiveness was compared with traditional techniques such as Convolutional Neural Network (CNN), Visual Geometry Group 16 (VGG-16), ResNet-50, Inception-V3 Efficient Net-B0. Using Brain Tumor MRI Dataset and training over 7022 high resolution MRI images in four classes: no tumor, glioma, meningioma and pituitary tumor. The results show the EffiRes-BT's superiority in terms of accuracy (0.9878), precision (0.9873), recall (0.9880) and F1-score (0-9879). Grad-CAM Visualisation highlights the models attention to tumor regions, with EffiRes-BT refining features in red colored regions and strengthens contextual learning in orange colored regions of heatmap. These results show the fundamental benefits of using the proposed EffiRes-BT model in medical imaging, especially for the brain tumor classification.