Principal Investigator: Rickmer Braren

Project goals

Artificial intelligence (AI) is increasingly being trained for specific tasks in the healthcare sector. For example, a network that detects the shrinkage of particular brain areas in dementia patients is unsuitable for analyzing the liver or heart. This leads to the need for many trained networks to be available for different tasks in the hospital. However, instead of keeping hundreds of such networks available locally, the data to be analyzed can be sent pseudonymously via a secure Internet connection to another hospital that maintains the required network. The data is analyzed there, the so-called inference is carried out, and the results are sent back via the same secure connection.

This necessary connection between hospitals, via which data can be exchanged quickly and securely, was already established at all participating sites during the first funding phase of the Medical Informatics Initiative (MII). The medical data integration centers (DIC) were set up for this purpose.

The OMI method platform aims to expand these DIZe with the ability to process medical image data. In a second step, the existing secure connections between the DIZs will be developed to transfer image data and integrate the on-site AI into this network. In this way, a hospital can use the AI of other hospitals without having to maintain it. 

The focus of our work is on:

• Specification of Open Medical Inference protocols.
• Development of the ImagingStudy part of the MII imaging extension module.
• Development of the DICOMweb™ adapter.
• Implementation of a reference gateway server and a reference client.
• These measures are intended to ensure that medical image data can be exchanged and analyzed efficiently and securely between hospitals, thereby significantly optimizing the use of AI in clinical practice.

Increasing efficiency through networked AI in medicine

A significant advance in healthcare informatics is developing a secure, distributed data exchange system based on the Medical Informatics Initiative (MII). This system enables the semantically interoperable exchange of multimodal health data for remote AI inference processes via open protocols and data formats. The research group is working on integrating basic OMI FHIR profiles, defining OMI conformance resources, and creating a comprehensive OMI implementation guide.

An essential part of this project is the reference implementation of an OMI client. This client is to be integrated into the local MII-compliant infrastructure of the participating partner locations. Tasks include:

• Integrating DIZ transfer services
• The continuous adaptation to agile development cycles
• Establishing an integration layer for consent and data usage policies
• Repeated performance and end-to-end testing

In addition, the user interface for controlling the OMI/DIZ interaction will be integrated, and extensive integration tests with DICOM and FHIR data will be performed.

Another critical element is the development and implementation of an OMI service register. This registry captures all running AI services in the OMI network, including their input/output data, availability, and usage metrics. The research group is designing a service architecture that ensures secure authentication, high availability, and fault tolerance. Automated tests of the gateway registration and client query interface ensure compliance with the OMI RESTful API and check performance in high-load scenarios.

Finally, all OMI components are rolled out to the partner sites, with each partner acting as a service provider and/or user. The platform is comprehensively evaluated by reviewing local infrastructure requirements, deploying service registers and clients, and customizing generic OMI reference gateways to the respective AI services. Algorithms and corresponding gateways are deployed at partner sites, and end-to-end testing is performed. These tests include functional validations, load, performance, latency tests, error handling tests, and disaster recovery tests.

This comprehensive approach aims to make the OMI system robust, interoperable, and optimally prepared for future medical applications, which will significantly improve the use of AI in clinical practice.