Improving Brain Tumor Detection Through Synthetic Data Augmentation using Conditional GANs

A brain tumor is one of the most dangerous disorders. A brain tumor, sometimes referred to as brain cancer, occurs when a group of cells within the brain turns cancerous and grows out of control, creating a mass. But early diagnosis can save the patient's life. For detection, we use magnetic resonance imaging (MRI). Although MRI reading and scanning are time-consuming, for doctors or specialists, the development of automated systems is often restricted by imbalanced and limited MRI datasets. This study explores a data augmentation framework using Conditional GANs (cGANs) to adjust balanced datasets for four classes: glioma, meningioma, pituitary tumors, and no-tumor controls. To assess synthetic data utility, five architectures-EfficientNetB0, InceptionV3, CNN+Swin Transformer, ConvNeXt Tiny, and MobileNetV3 Large-were trained exclusively on generative images and tested on a held-out realworld MRI test set. Our proposed model The hybrid CNN+Swin Transformer model, achieved the highest accuracy of 89.45 %, demonstrating superior performance in capturing both local features and global context. Despite Continuously challenges in differentiating glioma from meningioma due to overlapping visual characteristics, results show that cGAN-based augmentation is an effective solution for training robust models in data-constrained medical environments. However, the performance gap between synthetic and real-data training highlights the need for further clarification in generative realism to improve clinical generalization.