<-Introduction       Go to ToC       Definitions ->

Technical Specification: Artificial Intelligence for Health (MPAI-AIH) – Health Secure Platform (AIH-HSP) V1.1 specifies:

  1. The architecture of an end-to-end system called Health Secure Platform composed of:
    1. Front Ends that:
      1. Acquire Health Data.
      2. Process Health Data with AI Modules based on Neural Networks producing AI Health Data (AIH Data).
      3. License AIH Data to the Back End by attaching Model Licences to Health Data.
      4. Share their Neural Network Models with the Health Back End for Federated Learning purposes.
    2. A Back End that:
      1. Collects and stores AIH Data.
      2. Stores AIH Data Licences as Smart Contracts in a Blockchain.
      3. Processes AIH Data with AI Modules based on Neural Networks.
      4. Provides access to and enables Third-Party users to process AIH Data based on Smart Contracts.
      5. Collects Neural Network Models from Health Front Ends.
      6. Produces and distributes newer Neural Network Models by applying Federated Learning to the received Neural Network Models.
    3. A Blockchain that manages Smart Contracts, representing Licence Models.
  2. The AI Modules (AIM) required by the architecture.
  3. The Data Types exchanged by AIMs.
  4. Taxonomies for Data, Processing, and Actors of the Health Secure Platform.

AIH-HSP relies on the following MPAI standards:

  1. Technical Specification: Artificial Intelligence Framework (MPAI-AIF) V3.0 specifying the AI Framework (AIF) where Health Data is processed by AI Modules (AIM).
  2. Technical Specification: Process Instance Trust Framework (MPAI-PTF) V1.0 specifying the data structures, processes, and protocols enabling AIMs to establish, evaluate, and maintain trust in the AIF.
  3. Technical Specification: Data Types, Formats, and Qualifiers (MPAI-TFA) V1.5 specifying the Health Data Type Qualifiers enabling AIH-HSP to specify Data Types that are independent of specific formats.

An AIH-HSP implementation should provide the functionalities of the Technical Specification in compliance with the relevant applicable legal frameworks such as the European Union General Data Protection Regulation (GDPR) [6] or Artificial Intelligence Act [7].

Technical Specification: Artificial Intelligence for Health (MPAI-AIH) – Health Secure Platform (AIH-HSP) V1.0 has been developed by the AI Health Development Committee (AIH-DC) of Moving Picture, Audio, and Data Coding by Artificial Intelligence (MPAI), the international, unaffiliated, non-profit organisation developing standards for AI-based data coding.

What is new in V1.1

AIH-HSP V1.1 updates the AIM Metadata of V1.0 so that it can be used with the AIM Metadata of MPAI-AIF V3.0:

  • The AIM Metadata of every AIM conforms to the AIM Metadata schema of MPAI-AIF V3.0: it states the AIM Type it describes (its Header), numbers the Ports of the same Direction and Data Type, and, in a Composite AIM, connects Ports by Data Type and Port Number.
  • AIH Data Processing has Register Request and Register Response input and output Ports, as registration is performed on both the Health Front End and the Health Back End, and an AIH Data Process Response output Port, as the Health Front End has processing performed on the Health Back End. Its inputs that a Composite AIM does not use are optional.
  • Health Back End identifies its Auditing AIM as AIH-ADT, as the Auditing AIM itself does.
  • The identifiers of the AIMs and of the Data Types become V1.1.

The functions of the Health Secure Platform are those of V1.0.

MPAI may develop new Versions or new Technical Specifications whose scope is related to the AIH-HSP V1.1 Technical Specification.

<-Introduction       Go to ToC       Definitions ->