Smart hospital: achieving interoperability and raw data collection from medical devices in clinical routine

Eimo Martens, Hans-Ulrich Haase, Giulio Mastella, Andreas Henkel, Christoph Spinner, Franziska Hahn, Congyu Zou, Augusto Fava-Sanches, Julia Allescher, Daniel Heid, Elena Strauss, Melanie-Maria Maier, Mark Lachmann, Georg Schmidt, Dominik Westphal, Tobias Haufe, David Federle, Daniel Rueckert, Martin Boeker, Matthias Becker, Karl-Ludwig Laugwitz, Alexander Steger, Alexander MüllerFrontiers Digit. Health 2024 · DOI

Abstract

Modern clinical environments generate vast quantities of high-fidelity data from medical devices, yet this information often remains trapped in proprietary silos. This paper details the implementation of a "Smart Hospital" framework at the Technical University of Munich, which automates the collection of structured parameters and raw biosignals from over 115 medical devices. By utilizing an interoperable middleware system, the researchers successfully bridged the gap between medical hardware and hospital IT, creating a research-ready database of over 114,000 patients without manual data entry.

Introduction

Hospitals are increasingly data-rich, but the utility of this data is frequently hampered by fragmentation. Medical devices—such as ventilators, ECG machines, and ultrasound scanners—produce precise measurements and raw signals that are critical for both clinical care and advanced AI modeling. However, in many current clinical settings, data integration is limited to non-evaluable formats like PDFs or text reports.

High-quality data is generated by a wide array of medical devices, yet often remains siloed.

The lack of semantic interoperability—where different manufacturers use varying nomenclature for the same physiological parameters—further complicates data normalization. This study addresses these "media breaks" by developing a system capable of automated, large-scale data ingestion and standardization.

Methodology

The researchers developed and deployed a Cardiovascular Information System (CVIS) to serve as an interoperable bridge. The architecture was built upon several key pillars:

  1. Infrastructure: A cluster of eight virtual machines supported by 55TB of storage was established to manage the high volume of raw signal and image data.
  2. Hybrid Integration Strategy: The system utilized international standards such as HL7 for administrative data and DICOM for imaging. For devices lacking open standard support, custom interfaces were engineered to communicate with proprietary hardware, normalizing data into a unified SQL database.
  3. Semantic Mapping: To ensure machine readability, parameters were mapped to international coding systems, specifically SNOMED-CT and LOINC.
  4. Research Accessibility: A Fast Healthcare Interoperability Resources (FHIR) interface was implemented, allowing for the extraction of standardized data bundles for external research and AI training.

Results and Data Analysis

Over a two-year evaluation period (2020–2022), the system demonstrated robust performance in a high-volume clinical environment.

The framework achieves fully automated data integration into an interoperable database.

The system successfully captured data from 114,858 patients, including detailed medical records for over 50,000 individuals. Key metrics include:

  • 4.5 million technical readings, with an average of 28.5 parameters per examination.
  • 684,696 image and signal files, encompassing ECGs, X-rays, and ultrasound objects.
  • 10.2 million bidirectional HL7 messages processed, averaging 14,000 per day.

The automation eliminated manual transcription errors and significantly reduced the time required to prepare data for national and international research registries.

Discussion

The success of this framework underscores that middleware is a critical component for contemporary hospitals. Rather than waiting for universal adoption of open standards by manufacturers, hospitals can achieve interoperability by deploying systems that handle proprietary interfaces and normalize data in real-time.

Furthermore, the collection of raw biosignals—rather than just clinical summaries—provides a foundation for multimodal AI analysis. However, a significant "semantic gap" remains; current international codes (LOINC/SNOMED) do not yet cover the full breadth of parameters recorded by modern medical devices, representing a continued bottleneck for global data sharing.

Conclusion

This study proves the feasibility of fully automated, large-scale data integration in clinical routine. While currently focused on cardiovascular medicine, the framework provides a blueprint for other specialties. Future work will involve expanding semantic libraries and refining the FHIR-based repository to support increasingly complex, AI-driven "Smart Hospital" applications.

Summary updated

BibTeX
@article{martens2024smart,
  author = {Eimo Martens and Hans-Ulrich Haase and Giulio Mastella and Andreas Henkel and Christoph Spinner and Franziska Hahn and Congyu Zou and Augusto Fava-Sanches and Julia Allescher and Daniel Heid and Elena Strauss and Melanie-Maria Maier and Mark Lachmann and Georg Schmidt and Dominik Westphal and Tobias Haufe and David Federle and Daniel Rueckert and Martin Boeker and Matthias Becker and Karl-Ludwig Laugwitz and Alexander Steger and Alexander Müller},
  title = {{Smart hospital: achieving interoperability and raw data collection from medical devices in clinical routine}},
  journal = {Frontiers Digit. Health},
  year = {2024},
  doi = {10.3389/fdgth.2024.1341475},
}
Figure