federated learning combined ensemble aggregation for brain tumor classification in magnetic resonance imaging

In recent years, the use of deep learning techniques in medical imaging has shown promising results, particularly in the classification of brain tumors from magnetic resonance imaging (mri) scans. this article proposes an innovative approach that combines federated learning (fl) with convolutional neural networks (cnns) and ensemble aggregation to enhance the accuracy of mri brain tumor image classification. the proposed method utilizes cnn architectures that are fine-tuned on local datasets at different client sites. the results from these models are then aggregated using ensemble aggregation techniques at a central server and utilize a meta-learner to determine optimal weights for client models based on their validation performance, and aggregate model parameters using weighted averaging. next, the improved model weights are sent back to the clients for further training. this approach not only preserves data privacy but also enhances model robustness. experimental results demonstrate that the proposed method outperforms traditional centralized training methods, achieving higher accuracy and better generalization on unseen data.