Parkinson’s disease (PD) is a progressive neurodegenerative disorder that affects both motor and cognitive functions, making early and accurate diagnosis essential for timely intervention and improved patient outcomes. This study focuses on automated PD detection using advanced deep learning techniques applied to neuroimaging data, including MRI and PET scans. A comparative analysis was conducted between two transformer-based architectures, Vision Transformer (ViT) and Swin Transformer, to evaluate their effectiveness in extracting discriminative features for PD classification. Among the models, the Swin Transformer demonstrated superior accuracy and robustness due to its hierarchical feature extraction and efficient local–global representation learning. Despite promising results, challenges such as computational complexity and model generalization remain. To address these limitations, the proposed work introduces an enhanced framework that integrates hybrid transformer architectures with optimization strategies to improve accuracy while reducing computational demands. The goal is to develop a scalable, automated, and clinically relevant AI-driven diagnostic tool capable of supporting real-time, early-stage PD detection. This framework aims to contribute to reliable, efficient, and practical deployment of deep learning systems in healthcare settings.