LEAF: A Federated Learning-Aware Privacy-Preserving Framework for Healthcare Ecosystem

Abstract

Over the last decades, the healthcare industry has been revolutionized heavily, especially after the Covid-19 surge. Various artificial intelligence (AI) approaches have also been explored during this era for their applicability in healthcare. However, traditional AI techniques and algorithms are prone to overfitting with minimal robustness to unseen or untrained data. So, there is a need for new techniques which can overcome the issues mentioned earlier. Federated learning (FL) can help design specific AI services for the network of hospitals with less overfitting and more robust modules. However, with the inclusion of FL, the problem related to user privacy is the biggest challenge, making the use of FL in the real world a grand challenge. Most solutions presented in the literature used blockchain technology to mitigate the issues mentioned earlier. However, it prevents third-party systems from penetrating the decision process, but the network devices can access shared data. Moreover, blockchain implementation requires new paradigms and infrastructure with an additional overhead cost. Motivated by these facts, the paper presents a limited access encryption algorithm incorporating FL (LEAF) framework, i.e., an encryption technique that solves privacy issues with the help of edge-enabled AI models. The proposed LEAF framework preserves user privacy and minimizes overhead costs. The authors have evaluated the performance of the LEAF framework using extensive simulations and achieved superior results. The achieved accuracy of the proposed LEAF framework is 3% higher than that of the traditional centralized and FL-based systems without compromising user privacy. In the best scenario, the proposed framework’s encryption process also compresses the data size by 4–5 times.

Ähnliche Arbeiten