A Comparative Analysis of CNN and Vision Transformer Architectures for Brain Tumor Detection in MRI Scans

This study evaluates the performance of convolutional neural networks (CNNs) and Vision Transformers (ViTs) in classifying various brain MRI scans for the detection of tumors. Model series such as EfficientNet, ConvNeXt, ViT, and SwinTransformer were trained on a publicly available multiclass brain tumor dataset. To support experimentation and reproducibility, a custom GUI-based deep learning software was developed, enabling users to train models, configure parameters, apply data augmentation, monitor performance metrics, and generate diagnostic reports.