Federated Learning with Gramian Angular Fields for Privacy-Preserving ECG Classification on Heterogeneous IoT Devices

The recent study on federated learning for ECG classification marks a significant advancement in privacy-preserving healthcare technology. By utilizing Gramian Angular Fields to transform ECG signals into images, the framework allows for effective feature extraction through Convolutional Neural Networks. This approach was validated across heterogeneous IoT devices, including a server, a laptop, and a Raspberry Pi 4, demonstrating its versatility in real-world applications. The FL-GAF model achieved an impressive classification accuracy of 95.18%, outperforming traditional single-client methods in both accuracy and training time. This research highlights the potential of lightweight, privacy-preserving AI solutions in IoT healthcare monitoring, paving the way for smarter health systems that prioritize patient confidentiality while leveraging advanced machine learning techniques.