Open Medical Inference Protocol

Introduction

This is the Implementation Guide (IG) for the Open Medical Inference (OMI) protocol. This IG aims to provide guidance on how to implement inference algorithms such that they can be found, located and requested. To achieve this we rely on the FHIR Standard and its open, well described API.

Executive Summary

The OMI protocol defines data structures and interactions with the OMI-Registry. The following activities are supported by the protocol:

  • registration of algorithms,
  • search for inference services,
  • request inference services,

Figure 1 outlines these activities:

omi-protocol-architecture

Overview of activities: (1) registration: the service provider (upper) has an algorithm and hardware to process requests from external sources as a service. He provides information about his algorithm in the registry, adds information on how to access this service and gets approved by the registry. The service is now part of the registry and can be found by any user. (2) search and request: a service consumer (e.g. a hospital, but also a medical specialist in private practice) has pictures and additional informaton in his research Picture Archiving and Communication System (PACS) and wants to request a service outside of his institution. He searches for a service within the OMI registry. A service suitable for his data is provided by Hospital 1. He then sends a data bundle to the service provider and receives an asynchronous answer. (3) training an algorithm: a service consumer sends the links on where to access images to the API of Hospital 1. He receives an asynchronous answer provided by the service provider containing the results.

Structure

Authors of this Implementation Guide

  • Patrick Werner (MOLIT Institute)
  • Stefan Sigle (MOLIT Institute)
  • Kevin Kaufmes (MOLIT Institute)
  • Georg Mathes (MOLIT Institute)
  • Simon Schweizer (University of Heilbronn)
  • Liliana Caldeira (Institute for Diagnostic and Interventional Radiology University Hospital Cologne)
  • Rene Hosch (Institute for Artificial Intelligence in Medicine)
  • Martin Dyrba (German Center for Neurodegenerative Diseases (DZNE))

Contributors

The following persons were part of the workgroup that created this Implementation Guide.

Site Name
MOLIT Stefan Sigle, Patrick Werner, Christian Fegeler, Georg Mathes, Kevin Kaufmes
TUM Rickmer Braren, Martin Boeker, Philip Gouder, Peter Pallaoro
UKF Eric Frodl, Andreas Bucher, Marvin Schmidt
UKFr Elmar Kotter, Julius Wehrle, Shura-Roman Stump, Tobias Wieland
UKK Yuanbin Wang
UKEr Hans-Ulrich Prokosch, Thomas Ganslandt
UKHD Marco Nolden, Klaus Meier-Hein
HHN Christian Fegeler, Simon Schweizer
UME Felix Nensa, Obi Pelka
UMR Mathias Manzke, Petra Gröber

Funding information

This IG is part of the work package 1 of the OMI project and was provided by MOLIT Institute. This project is funded by the BMBF (funding code: 01ZZ2315J). For further information please review the following link