Multi-Continental Healthcare Modelling Using Blockchain-Enabled Federated Learning

TL;DR

This paper introduces a blockchain-enabled federated learning framework for privacy-preserving, cross-continental health modelling, using glucose prediction as a case study. The Multi-Continental Glucose Prediction (MCGP) framework integrates FL with an on-chain incentive mechanism to reward honest participants and deter malicious actors, while enabling off-chain aggregation to reduce costs. Experimental results across five hospitals and simulated 30 T1D patients show that MCGP outperforms single-site and, in some cases, centralized training, with strong generalization to unseen patients and resilience to adversarial participants. The work demonstrates the feasibility and benefits of international collaboration in healthcare data analytics without sharing raw data, signaling a step toward bias reduction and broader disease modelling.

Abstract

One of the biggest challenges of building artificial intelligence (AI) model in the healthcare area is the data sharing. Since healthcare data is private, sensitive, and heterogeneous, collecting sufficient data for modelling is exhausting, costly, and sometimes impossible. In this paper, we propose a framework for global healthcare modelling using datasets from multi-continents (Europe, North America, and Asia) without sharing the local datasets, and choose glucose management as a study model to verify its effectiveness. Technically, blockchain-enabled federated learning is implemented with adaptation to meet the privacy and safety requirements of healthcare data, meanwhile, it rewards honest participation and penalizes malicious activities using its on-chain incentive mechanism. Experimental results show that the proposed framework is effective, efficient, and privacy-preserving. Its prediction accuracy consistently outperforms models trained on limited personal data and achieves comparable or even slightly better results than centralized training in certain scenarios, all while preserving data privacy. This work paves the way for international collaborations on healthcare projects, where additional data is crucial for reducing bias and providing benefits to humanity.

Figures (5)

• Figure 1: A diagram that illustrates blood glucose prediction modelling without sharing private data for five participants from three continents.
• Figure 2: Overview of the MCGP framework architecture. Participants (hospitals) are randomly assigned roles as either proposers or voters to collaboratively train a model using a reward and slashing mechanism. Malicious participants are identified during training, assigned no role, and removed from the FL training task. The final optimized global model is then deployed to test on unseen patients’ data.
• Figure 3: Incentive Mechanism of BCFL.
• Figure 4: LSTM model test results on Patient23’s local data for the last 1000 points over a three-week period, with glucose levels measured in mg/dL. To better display the values, the glucose levels have been scaled down to half of their original size. The closer to the Ground Truth means the better the result.
• Figure 5: LSTM model test results on unseen data from Patient 30 over the last 1000 points in a three-week period. Close to Ground Truth is better.

Theorems & Definitions (3)

• Definition 1
• Definition 2
• Definition 3