Brain Tumor Classification Based on Federated Learning

This paper investigates the challenges of brain tumor classification and focuses on diagnostic performance while preserving the patient's privacy by investigating decentralized using federated learning. Brain tumor are considered one of the significant health risks and a widespread cancer disease, and the more accurate classification methods will help to diagnose the disease and for effective treatment planning. Conventional methods encounter many difficulties because of the limited availability of diverse medical imaging data as well as privacy regulations. To address these challenges, this method allows for decentralised training across multiple data centres by using the Federated Learning method. The proposed method utilises a pre-trained DenseNet model within an FL environment. This approach guarantees effective feature selection, improving the overall performance of the classification model. The training data remains localised at each node, and only the trainable weights and model updates are shared, therefore preserving data confidentiality. The FL model collects these updates to build a model that is capable of classifying MRI images. The study assesses the efficacy of the FL model through the utilisation of three publicly accessible MRI datasets, thereby substantiating a notable enhancement in classification accuracy when compared to single-institution models. The results indicate that the FL approach not only enhances diagnostic accuracy but also facilitates multi-institutional collaborations without compromising patient data privacy. This solution holds promise for widespread clinical adoption, enabling better management and treatment of brain tumors through advanced, privacy-preserving AI techniques.