This paper presents a novel approach for multi-modal medical image fusion and disease classification using a Transformer-GAN network. The framework integrates transformer models for feature extraction and attention-based fusion with GANs to generate high-quality fused images, ensuring both spatial and semantic consistency. The system also addresses image registration challenges by aligning images from different modalities, enhancing fusion accuracy. Experimental results demonstrate that proposed network outperforms existing methods in key metrics like structural similarity, mutual information, and classification accuracy. The fused images improve brain tumor diagnosis by providing a comprehensive view of complementary modalities such as MRI, CT, and PET scans. This hybrid framework also shows promising prospects for real-time applications, offering faster processing with high precision. The combination of transformer models and GANs bridges the gap between high-quality image fusion and robust disease classification, making it a valuable tool in medical diagnostics.