Brain tumor classification via MRI scans poses challenges related to privacy, data decentralization, and model transparency. This work proposes a secure, decentralized pipeline combining Federated Learning (FL) with blockchain-based ver-ification for multi-class tumor detection. Using ResNet-SO with transfer learning, the model distinguishes glioma, meningioma, and pituitary tumors from colorized MRI images. Grayscale scans are enhanced using histogram equalization and edge sharpening to align with RGB-pretrained networks. Training is distributed across simulated hospital nodes via the Flower FL framework, leveraging Federated Batch Normalization (FedBN) to address non-IID data. A smart contract on Ethereum’s Sepolia testnet records model hashes, accuracy, and timestamps for auditability. The approach achieves 97.98 % global accuracy, outperforming CNN baselines. This system offers a scalable, transparent, and privacy-preserving solution for AI-driven med-ical imaging.