Vision Transformers have been extensively used in natural language processing tasks; nevertheless, their application in medical imaging remains largely unexamined. This research examines ADHD classification with pre-trained vision transformers, including Tiny-ViT, Mobile-ViT, and ViT, and evaluates their effectiveness in this job. We also want to develop a pipeline for preprocessing fMRI pictures that is less resource-intensive and more efficient. This work involves ADHD-200 dataset with 500 subjects. Established a preprocessing pipeline, tried 6 models, which included VGG-19, ViT Tiny, Mobile ViT, ViT, 3D CNN + ViT hybrid models, and compared them, achieving 87% accuracy.