This paper proposes a federated learning framework that enables the training of machine learning models utilizing structural brain imaging data across decentralized devices while maintaining data privacy. By using Transfer Learning, a global model is defined in a federated learning setting to provide more dependable and accurate Alzheimer’s disease diagnosis. The aggregated model which is a result of this framework, achieves a classification accuracy of 94.64%. Explainable AI techniques like Grad-CAM visualizations are used to enhance the transparency and interpretability of the model trained on diverse client datasets. The resulting framework demonstrates the potential of federated transfer learning, promoting model training for Alzheimer’s disease detection without sharing raw data while also being scalable for other neuroimaging datasets.