Communication-Efficient Federated Learning (CEFL) for CT Image Classification in Bandwidth-Constrained Wireless Healthcare Networks

The increasing adoption of deep learning for Computed Tomography (CT) image classification has significantly improved diagnostic accuracy in medical imaging. However, traditional centralized approaches require transferring large volumes of medical data to a central server, leading to high bandwidth consumption, increased latency, and serious privacy concerns, particularly in wireless healthcare environments. Federated Learning (FL) offers a promising solution by enabling collaborative model training without sharing raw patient data. Nevertheless, conventional FL methods suffer from substantial communication overhead due to frequent transmission of large model updates, limiting their applicability in bandwidth-constrained networks. To address these challenges, this paper proposes a Communication-Efficient Federated Learning (CEFL) framework for distributed CT image classification. The proposed approach integrates gradient sparsification, model quantization, and adaptive communication scheduling to significantly reduce the size and frequency of model updates. The framework is implemented using a multi-layer architecture comprising medical imaging, edge computing, wireless communication, and federated aggregation layers. Experiments are conducted on the LIDC-IDRI CT dataset under simulated bandwidth-constrained conditions. The results demonstrate that the proposed CEFL framework reduces communication overhead by up to 40 - 60 compared to conventional FL methods such as FedAvg, while achieving improved classification accuracy of approximately 90. Furthermore, latency is significantly reduced, making the system suitable for real-time wireless healthcare applications. These findings highlight the effectiveness of communication-efficient strategies in enabling scalable, privacy-preserving medical image analysis.