Multimodal based Brain Tumor Classification using Clustered Federated Learning in Edge based IoT Network

A brain tumor is characterized by the uncontrolled growth of cells that causes a severe impact on the quality of life. Nowadays, various methods based on multimodal medical images are used to identify brain tumors. Meanwhile, these methods require extensive amounts of labeled information to identify disease tissues and often encounter various challenges based on privacy and accuracy. This paper establishes a Taylor-based Focal Log-cosh Entropy Network (TFLE-Net) to classify brain tumors from demographic data and Magnetic Resonance Imaging (MRI) images by utilizing clustered federated learning in an edge-based Internet of Things (IoT) network. The detection process is executed in clustered federated learning concerning two entities: server and nodes. In a local training model, an input MRI scan is pre-processed, further brain tumor areas are segmented, and then, the mining of various features is performed. Simultaneously, the normalization of input demographic data is done, and the suitable features are selected using information gain. Further, the extracted features and chosen features are subjected to TFLE-Net to classify brain tumors. Additionally, the TFLE-Net yielded Mean Absolute Error (MAE) of 0.057, Root Mean Square Error (RMSE) of 0.047, True Positive Rate (TPR) of 97.789%, and accuracy of 96.987%.