Attention-Deficit Hyperactivity Disorder (ADHD) represents a challenging neurodevelopmental disorder that consistently displays three major symptoms involving inattention and hyperactivity alongside impulsivity. Traditional approaches for diagnosis use behavioral evaluations that create both wrong conclusions and delayed help timing. This research develops a complete diagnostic solution involving deep learning federated learning and blockchain security to analyze actigraphy signals originating from IoMT devices. This method first uses UMAP as well as PCA and t-SNE to reduce data dimensions before implementing a hybrid CNN-Transformer neural network to achieve improved classification results. A distributed learning method helps medical institutions run model training autonomously while satisfying privacy rules and addressing data centralization challenges. Model updates on blockchain systems gain protection through smart contracts and cryptographic hashing to stop adversarial attacks and sustain data authenticity. Laboratory tests reveal that this approach reaches 99.2% classification precision without significant performance impact, establishing its effectiveness. This presented study provides on-the-next level ADHD diagnosis features with the help of an AIbased system that ensures privacy and guarantees tampering and scalable operations. Such results allow advancing accurate medical works by real-time monitoring of ADHD and offer safe application of medical Artificial Intelligence to distributed healthcare processes. This will provide objective and credible evaluations that will exist on a global scale.