Multi-Class Brain Tumor Classification of MRI Image Using Federated Learning with Blockchain

Brain tumors represent a critical medical concern, accounting for a significant number of mortalities annually. Early and accurate diagnosis is paramount for guiding effective treatment strategies. However, the scarcity of diverse data poses a challenge in constructing an accurate global model. Most healthcare institutions are often reluctant to share patient data due to privacy and legal concerns. In this situation, federated learning emerges as a viable solution, addressing privacy and data diversity concerns. In this work, we propose each institution will train a deep learning classifier model using their local data. Then, while keeping patient data private, only local model parameters are shared with the global model. The proposed system utilizes Blockchain to ensure secure transfer of model parameters. Our model demonstrates exceptional accuracy in classifying four distinct tumor types, showcasing the potential of this framework for collaborative medical diagnosis while preserving patient data privacy. During aggregation, we employ a weighted average technique to enhance accuracy. Additionally, we explore an acceleration method, significantly expediting the training phase of the global model. The accelerated weighted average aggregation technique yields an impressive 93.75% accuracy for the global model.