Federated Learning-Based Privacy-Preserving ECG Signal Classification for IoMT

With the rapid development of the Internet of Medical Things (IoMT) and deep learning technology, automatic electrocardiogram (ECG) signal classification has become increasingly important for real-time health monitoring. However, the privacy and security of sensitive medical data remain critical challenges. Achieving efficient and accurate signal classification while ensuring data confidentiality and enhancing model interpretability is a pressing issue in current research. To address these concerns, this paper proposes a Federated Learning (FL)-based ECG signal classification framework incorporating privacy-preserving mechanisms to ensure the security of client data. The proposed approach integrates Differential Privacy and Secure Aggregation, and is designed to handle practical deployment challenges, such as client dropout. Given that noise injection techniques may increase communication overhead, a parameterefficient federated fine-tuning mechanism is introduced to reduce the amount of data transmitted during training. In addition, Explainable Artificial Intelligence (XAI) techniques are applied to visualize and interpret the classification results, thereby enhancing model transparency and interpretability. Experimental results demonstrate that the proposed method achieves a classification accuracy of $\mathbf{9 8. 1 4 \%}$, comparable to state-of-the-art FL algorithms, while also effectively preserving user privacy and minimizing communication and computational overhead.