Abstract
Artificial Intelligence (AI) is increasingly used in healthcare to support the processing,interpretation, and analysis of medical data. While AI-based systems can improve clin
ical workflows, their use also introduces challenges related to transparency, traceability,
accountability, and the trustworthy handling of patient data. In clinical contexts, docu
menting only the final result is often insufficient. It should also be possible to reconstruct
which AI system was used, which data contributed to an output, under which conditions
the system was executed, and how the result was reviewed by a healthcare professional.
This thesis addresses the gap between high-level European regulatory transparency
requirements and their concrete technical representation in interoperable healthcare
information systems. The aim is to develop a conceptual model based on Fast Health
care Interoperability Resources (FHIR) for the transparent documentation of AI usage
in medical data processing. Based on a qualitative Design Science Research approach,
transparency- and documentation-related requirements are derived from the European
Union Artificial Intelligence Act (AI Act), the General Data Protection Regulation
(GDPR), and the European Health Data Space (EHDS). These requirements are trans
lated into a conceptual model and mapped to HL7 FHIR Release 5 through profiles,
extensions, terminology bindings, and an Implementation Guide.
The proposed model structures AI-related transparency information into three con
nected contexts: static AI system metadata, patient-specific AI output context, and
clinical decision context. This structure enables the documentation of system identifi
cation, execution metadata, data provenance, legal context, audit information, human
oversight, and patient-facing transparency. A lightweight proof of concept based on a
synthetic acute-care scenario demonstrates that representative AI-related metadata can
be transformed into FHIR resources and syntactically validated against the defined
profiles.
The results show that the proposed model provides a technically feasible structure for
supporting the traceability and auditability of AI-supported medical data processing.
However, the validation is limited to syntactic conformance and does not constitute
clinical validation, legal compliance certification, or production readiness.
| Date of Award | 2026 |
|---|---|
| Original language | English (American) |
| Awarding Institution |
|
| Supervisor | Gerald Adam Zwettler (Supervisor) |
Studyprogram
- Data Science and Engineering
Cite this
- Standard