<p>The paper focuses on the improved brain Magnetic Resonance Imaging (MRI) image classification using a complementary method that integrates federated learning with the multi-model ensemble approach. The data employed here include 7023 brain MRI images divided into four classes: glioma, meningioma, no tumor, and pituitary tumors. It is a blend of the top three convolutional neural networks (CNNs) VGG16, ResNet50, and InceptionV3 with the top features of each model into a new ensemble architecture to improve feature extraction capacity. The most important changes are selective freezing of layers: best transfer learning from pre-trained networks in order not to add extra parameters, and data variability augmentation: increase data variability for improved robustness against MRI scan variation. Federated learning was added to leverage decentralized data, and this improves privacy and data security, a core necessity in medical use. The methodology provided significant improvements in classification metrics with 98% accuracy and other resilience performance metrics. Edge AI is applied in the study by integrating federated learning and multi-model ensembles to enhance privacy and computational efficiency for brain MRI classification. The Edge AI solution facilitates decentralized data processing, which is imperative when dealing with sensitive medical information, with robust security features.</p>