A Solution for the Health Data Sharing Dilemma: Data-Less and Identity-Less Model Sharing Through Federated Learning and Digital Twin-Assisted Clinical Decision Making

Digital twins are essentially digital replicas of physical entities. Their usage is becoming more common across various industries, including healthcare. However, the implementation of digital twins in healthcare is uniquely challenging. This is partly because of the sensitive nature of health data and privacy concerns. These concerns limit health data accessibility and shareability. This paper attempts to address this challenge of health data sharing. We propose a novel approach that leverages federated learning, model sharing, and digital twin-assisted clinical decision making. Our approach ensures that health data are kept federated with healthcare providers. Healthcare providers train machine learning models on their own data. Then, instead of sharing the data, the trained models are shared. This is enabled via an arrangement like a private blockchain that is accessible to subscribed healthcare providers. This approach allows healthcare providers to access and use machine learning models for clinical decision support without compromising sensitive data about patients. Certain information about machine learning models will be shared. These include indicators such as the sample size on which a model has been trained on, validation metrics, and model accuracy. Such information assists other healthcare providers in selecting the most effective models. We demonstrate the efficacy of this approach through a case study on chronic disease management (e.g., cancer) using Liquid Neural Networks. Our results show how federated learning and model sharing can enhance clinical decision making and improve patient outcomes while ensuring the privacy of data.