<-References     Go to ToC     AI Workflows->

(Informative)

1     Architecture 2    Operation 3. Services

1        Architecture

The Health Secure Platform (HSP) is a system composed of a Health Back End (HBE), a plurality of Health Front Ends (HFE), and a Blockchain. End Users and Third-Party Users access the Health Secure Functions as describes in the Operation Section.

Figure 1 graphically depicts the elements of the AI for Health – Secure Platform.

Figure 1 – General Model of AIH-HSP V1.0

2         Operation

The operation of the HSP is best described by this workflow:

  1. An End User
    1. Is equipped with device running an AIF-enabled app – Health Front End (HFE).
    2. Acquires and uniquely identify Health Data.
    3. Creates AIH Data by combining Health Data with a Model Licence and uniquely identifying it.
    4. Processes AIH Data using AI Modules (AIM) downloaded from the MPAI Store. Neural Networks in the AIMs continually learn while making inferences on AIH Data.
    5. Creates a new uniquely identified AIH Data  instance.
    6. Uploads un-processed and processed AIH Data to the Health Back End
  2. The Health Back End
    1. Downloads the latest AIMs from the MPAI Store.
    2. Stores the Model Licence as a Smart Contract on a Blockchain.
    3. Converts the Model Licence into a Smart Contract.
    4. Add the ID of the Smart Contract to the AIH Data.
    5. Processes the AIH Data in the internal AI Framework as permitted by the terms and conditions expressed in the Smart Contract.
    6. From time to time sends a request to a class of End Users to receive the trained Neural Network Model with a specific untrained Model ID
  3. Requested End Users send their trained Neural Network Models.
  4. Health Back End
    1. Produces a new Neural Network Model from the received trained Models using Federated Learning techniques.
    2. Uploads the new Neural Network Model to the MPAI Store.
    3. Notifies End Users that the new Neural Network Model is available at the MPAI Store.
  5. A Third-Party User belonging to one of the categories identified in the AI Health Taxonomy
    1. Registers with the Health Back End.
    2. Processes the AIH Data in the internal AI Framework as permitted by the terms and conditions expressed in the Smart Contract.
    3. Uses the processes AIH Data as permitted by the terms and conditions expressed in the Smart Contract.

The AIH Taxonomy identifies:

  1. Users: currently: End User, Non-Profit Entity, Profit Entity, Clinical Entity, Authorised Entity, Caregiver.
  2. AIH Data
    1. Classes (currently: ECG, EEG, Genomics, and Medical Images).
    2. Statuses (currently, Anonymised, Pseudonymised, Identified)
    3. Usages (currently, Unrestricted, Pseudonymised, Anonymised, Research, Patient use, Health care)
    4. Processing Types (currently: ECG, EEG, Genomics, and Medical Images).
  3. Algorithms:
    1. Anonymisation
    2. De-Identification
  4. Anomalies:  Types.

Note that the Operation of an implementation of an AIF instance is required to be Zero Trust. Technical Specification: AI Framework (MPAI-AIF) V3.0 provides a set of requirements that a Zero-Trust implementation of an AIF instance is expected to satisfy.

3        Services

The Health Secure Platform is composed of a set of distributed components and services:

  1. The Front End, the End User’s personal gateway to their external biometric sensors and any AIH Data that:
    1. Captures End User’s Health Data, e.g., from Google Fit and Apple Health, and external biometric sensors that capture Health Data.
    2. Locally stores AIH Data in a “Secure Data Vault” controlled by the End User.
    3. AI processes AIH Data using standard AIMs and AIWs downloaded from the MPAI-Store performing the computational operations on the End User’s AIH Data, including transformations, training, and inferences.
    4. Alerts the End-User about any deviation of the value of the AIH Data that may be caused, e.g., by disease, injury, or chronic conditions.
    5. Uploads the processed AIH Data to the Back End.
  2. The AIH Back End, composed of a set of tools that implement the necessary services
    1. Securely stores, de-identifies and anonymises AIH Data, controls entity authentication and access to data, and licenses and audits the access to Back End AIH Data.
    2. Gathers anonymised data from End Users and acts as a broker gateway between Third-Part Entities requesting access to AIH Data and its providers.
    3. Grants access rights without referring to the identity of the End Users providing the data. The Back End may only grant the Third-Party User the rights to process AIH Data that the Back End has been specifically granted by the relevant End User.
  3. Blockchain enables the system’s transparency and auditability. Each provision of and access to AIH Data requires the emission of a license in the form of a Smart Contract that is stored on the Blockchain. The Smart Contract contains information about:
    1. The parties, e.g., the End User sending AIH Data and the Back End, and any future Third-Party User requesting access to and processing AIH Data.
    2. The Type of Third-Party User (per the MPAI-AIH Taxonomy).
    3. The AIH Data and AIH Models to be used.
    4. The Rights granted to use the AIH Data:
      1. Type of use of the AIH Data (per the MPAI-AIH Taxonomy).
      2.  Type of use of the processed AIH Data (per the MPAI-AIH Taxonomy).
    5. The duration of the Licence.
  4. The AI Services offered by the Back End can be used directly to process the AIH Data on the Front End and extract the specific knowledge sought by the End User or Third-Party Users based on the Licence. These services are selected from those available from the MPAI Store and may be orchestrated to produce specific analyses for the Third-Party Users that request access to AIH Data. By means of data processing, AI services enable specific and customised training of Machine Learning Models to identify and assist in the identification of medical diagnosis and prognosis.
  5. The AI Federated Learning System (FLS) orchestrates the learning of a central model for medical diagnosis and prognosis, namely by working as a medical anomaly detection tool, receiving Neural Network Model weights data from the Front End and using it under the terms of the Smart Contract that was established between the End User and the Back End. When an improved model is obtained by the FLS, this is uploaded to the MPAI-Store.

<-References     Go to ToC     AI Workflows->