Cutting Edge Deep Learning Models for Brain Tumor Classification

Accurate differentiation of brain tumor types is essential for effective treatment planning, yet manual interpretation of MRI scans is labor-intensive and prone to variability. This study presents a systematic benchmark of ten state-of-the-art deep learning models for automated brain tumor classification using a public dataset of 3,064 contrast-enhanced MRI scans covering glioma, meningioma, and pituitary tumors. Five advanced Convolutional Neural Networks (CNNs), including Inception v4 and ConvNeXt, are directly compared against five Vision Transformer (ViT) architectures, such as Swin-Base and PiT-Base. Models were evaluated under a unified framework with transfer learning, data augmentation, and macro-averaged metrics (accuracy, precision, recall, F1-score). Results show that Inception v4 achieved the highest accuracy (96.73%) among CNNs, while PiT-Base attained a competitive accuracy of 96.41% with fewer parameters, highlighting the trade-off between multi-scale convolutional processing and efficient transformer designs. Explainable AI analysis with Grad-CAM confirmed that models consistently focused on clinically relevant tumor regions, enhancing interpretability. The findings underscore that both CNNs and ViTs can reach near-expert diagnostic performance, with architectural efficiency rather than parameter count being the decisive factor.