Blockchain-Based Two-Stage Federated Learning With Non-IID Data in IoMT System

• Published in: IEEE transactions on computational social systems Vol. 10; no. 4; pp. 1 - 10
• Main Authors: Lian, Zhuotao, Zeng, Qingkui, Wang, Weizheng, Gadekallu, Thippa Reddy, Su, Chunhua
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.08.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial intelligence; Blockchain; Blockchains; Cryptography; Data models; Data privacy; Federated learning; Internet of medical things; Internet of Medical Things (IoMT); Machine learning; Model accuracy; non-independent identically distributed (non-IID) data; Privacy; privacy preservation; Security; Servers; Training; Wireless communications
• ISSN: 2329-924X; 2373-7476
• DOI: 10.1109/TCSS.2022.3216802

The Internet of Medical Things (IoMT) has a bright future with the development of smart mobile devices. Information technology is also leading changes in the healthcare industry. IoMT devices can detect patient signs and provide treatment guidance and even instant diagnoses through technologies, such as artificial intelligence (AI) and wireless communication. However, conventional centralized machine learning approaches are often difficult to apply within IoMT devices because of the difficulty of large-scale collection of patient data and the potential risk of privacy breaches. Therefore, we propose a blockchain-based two-stage federated learning approach that allows IoMT devices to train a global model collaboratively without gathering the data to a central server. Specifically, to address the problem of poor training performance on non-independent identically distributed (non-IID) data, we design a blockchain-based data-sharing scheme that can significantly improve the model's accuracy without threatening user privacy. We also design a client selection mechanism to further improve the system's efficiency. Finally, we validate the feasibility and effectiveness of our system through simulation experiments on three popular datasets (i.e., MNIST, Fashion-MNIST, and CIFAR-10).