[Paper Review] Federated Learning for COVID-19 Detection with Generative Adversarial Networks in Edge Cloud Computing

Dinh C. Nguyen, Ming Ding+3 more|arXiv (Cornell University)|Oct 13, 2021

COVID-19 diagnosis using AI44 references135 citations

TL;DR

This paper proposes FedGAN, a federated learning and generative adversarial network (GAN) framework for privacy-preserving COVID-19 detection in edge cloud environments. By training local GANs at hospitals and aggregating model updates via a decentralized blockchain-based FL system with differential privacy, FedGAN generates realistic synthetic X-ray images without sharing real data, achieving high detection accuracy with reduced training latency compared to state-of-the-art methods.

ABSTRACT

COVID-19 has spread rapidly across the globe and become a deadly pandemic. Recently, many artificial intelligence-based approaches have been used for COVID-19 detection, but they often require public data sharing with cloud datacentres and thus remain privacy concerns. This paper proposes a new federated learning scheme, called FedGAN, to generate realistic COVID-19 images for facilitating privacy-enhanced COVID-19 detection with generative adversarial networks (GANs) in edge cloud computing. Particularly, we first propose a GAN where a discriminator and a generator based on convolutional neural networks (CNNs) at each edge-based medical institution alternatively are trained to mimic the real COVID-19 data distribution. Then, we propose a new federated learning solution which allows local GANs to collaborate and exchange learned parameters with a cloud server, aiming to enrich the global GAN model for generating realistic COVID-19 images without the need for sharing actual data. To enhance the privacy in federated COVID-19 data analytics, we integrate a differential privacy solution at each hospital institution. Moreover, we propose a new blockchain-based FedGAN framework for secure COVID-19 data analytics, by decentralizing the FL process with a new mining solution for low running latency. Simulations results demonstrate the superiority of our approach for COVID-19 detection over the state-of-the-art schemes.

Motivation & Objective

• Address the challenge of limited and privacy-restricted access to distributed COVID-19 X-ray datasets across hospitals.
• Overcome data privacy concerns in centralized AI training by avoiding actual data sharing.
• Improve the quality and diversity of training data for COVID-19 detection using synthetic images generated via GANs.
• Ensure secure and reliable federated model aggregation by replacing central servers with a decentralized blockchain-based framework.
• Minimize training latency in federated learning through a novel, lightweight proof-of-reserve (PoR) consensus mechanism for blockchain.

Proposed method

• Train a local GAN at each hospital using convolutional neural networks (CNNs) for generator and discriminator, mimicking real COVID-19 X-ray data distribution.
• Implement federated learning to aggregate local model parameters from hospitals at a central cloud server, avoiding raw data transmission.
• Integrate differential privacy at each hospital by adding calibrated noise to model updates to protect individual data.
• Propose a blockchain-based FedGAN framework using a novel proof-of-reserve (PoR) consensus mechanism to decentralize model aggregation and reduce latency.
• Design a lightweight block verification strategy in PoR to minimize mining latency, especially for large data blocks.
• Use a hybrid edge-cloud architecture where edge nodes (hospitals) train local GANs and communicate only model updates to the cloud or blockchain network.

Experimental results

Research questions

• RQ1 Can a federated GAN framework effectively generate high-quality synthetic COVID-19 X-ray images without sharing real patient data?
• RQ2 How does integrating differential privacy into federated GAN training affect model utility and privacy guarantees?
• RQ3 Can a blockchain-based consensus mechanism reduce latency in federated learning while ensuring security and decentralization?
• RQ4 What is the impact of a novel PoR consensus mechanism on block verification latency compared to traditional DPoS in federated medical AI?
• RQ5 How does the performance of the proposed blockchain-based FedGAN compare to centralized and non-blockchain federated baselines in terms of detection accuracy and training efficiency?

Key findings

• The proposed blockchain-based FedGAN scheme achieves competitive detection accuracy with the baseline FedGAN, particularly on the DarkCOVID dataset, demonstrating that decentralization does not degrade model performance.
• The PoR consensus mechanism reduces block verification latency by up to 23.2% on the DarkCOVID dataset and 19.5% on the ChestCOVID dataset compared to traditional DPoS-based blockchain FL.
• The integration of differential privacy at each hospital significantly enhances data privacy without compromising the quality of the generated synthetic images or the detection accuracy.
• The federated GAN framework reduces the need for large-scale data collection by generating realistic synthetic X-ray images that improve training performance across distributed institutions.
• The overall running latency of the proposed blockchain-based FedGAN is significantly lower than traditional blockchain-FL schemes, making it suitable for real-time medical AI applications.
• The simulation results confirm that FedGAN outperforms state-of-the-art schemes in both detection accuracy and training efficiency, especially under privacy-constrained and decentralized settings.