Personal Health Train

Distributed Privacy-Preserving Data Analysis

Background

In public health research, data sharing is often challenging due to the sensitive nature and heterogeneous characteristics of health data. Moreover, the lack of standardization and semantics further exacerbate the problems of data fragments and data silos, which makes data analytics difficult. While there are great benefits for collaborative analyses across multiple epidemiological studies, data protection concerns and the lack of IT infrastructure currently limit widespread cross-institute research projects. The aim of NFDI4Health is to enable institutes to participate in such

research projects without actually ceding control over their data. One approach to overcome these challenges is federated data analysis with Personal Health Train (PHT). The PHT complements the Central Search Hub (CSH) and Local Data Hub (LDH) by providing the means of executing analysis in a distributed and privacy-preserving manner. Unlike centralized analysis, the PHT brings the algorithm to the data instead of vice versa, enabling data owners to retain control of their data and keeping the data in its origin.

Concept

The Personal Health Train Concept is an innovative approach to foster data-driven innovations in medicine. It provides a distributed analysis infrastructure that enables research on sensitive data without prior data sharing, while supporting diverse data formats. The PHT originates from an analogy from the real world, resembling a railway system with trains, stations, and train depots. In the PHT ecosystem, the train encapsulates an analytical task, represented by the goods in the analogy. The data provider plays the role of a reachable station, accessed by the train. The station executes the task, processing the available data. The Central Service (CS)

serves as the depot, managing train orchestration, operational logic, business logic, and data management. This design paradigm ensures algorithms are brought to the data instead of bringing confidential data to the algorithm, ensuring compliance with data protection requirements. Hereby, the PHT provides a distributed, flexible approach to using data in a network of participants, incorporating the FAIR principles. Within Germany different implementation initiatives such as PADME or PHT-meDIC are cooperating closely as part of the international PHT Go FAIR implementation network.

Implementation under NFDI4Health

To demonstrate the effectiveness of the NFDI4Health infrastructure, we conducted 2 Use Cases in collaboration with the University Hospital Cologne Radiology Department (UHC) and Fraunhofer MEVIS. Hereby, the PHT is used to generate synthetic data in a distributed manner, which can later aid in data harmonization efforts. The Use Case developed with Fraunhofer MEVIS focuses on the recognition of kidney tumors in patients using

computer tomography images from 2 different locations, where known tumor patients have already been treated. With this data, our Use Case aims at studying the overall outcome of different therapy approaches. In the Use Case conducted with the UHC Radiology Department, we targeted at the recognition of lung-cancer patients, where data-harmonization is a pressing issue due to many different devices and protocols found in practice.