P2FL: Privacy-Preserving Federated Learning Approach for Healthcare Informatics at the Edge

The healthcare industry is at significant risk of cybercrime and privacy violations due to the extensive distribution and sensitivity of health data. Recent trends in confidentiality breaches across multiple industries highlight the critical need for improved data security technologies that ensure privacy, accuracy, and reliability. Furthermore, decentralized healthcare systems suffer from intermittent remote clients and imbalanced data. This paper proposes a privacy-preserving federated learning (P2FL) approach for improving medical data privacy in health care informatics on edge devices. The remote clients in FL may have imbalanced datasets for local training, leading to lower results. Data augmentation procedures are employed in local model training to address this issue and balance datasets. Dynamic customer interactions with the global server via remote healthcare facilities are challenging. In real-world scenarios, technical or connectivity issues may cause clients to attend or exit the training session. The proposed methodology is tested on various clients and image sizes to evaluate its efficacy under various conditions. The suggested method achieved a 99.04% classification accuracy using two standard datasets. These findings enable collaborative efforts among medical institutions to leverage private data efficiently, further developing strong patient diagnostic models.