<-AI Modules     Go to ToC        Datasets->

1 Definition 2 Functional Requirements 3 Syntax 4 Semantics

1 Definition

Taxonomies of AIH Data.

2 Functional Requirements

Taxonomies cover:

  1. AIH Data Classes
  2. AIH Data Users
  3. AIH Data Statuses
  4. AIH Data Usages
  5. Anonymisation/De-Identification Algorithms
  6. Anomaly Types

3 Syntax

https://schemas.mpai.community/AIH1/V1.0/data/AIHTaxonomies.json

4 Semantics

Label Description
Header AIH Taxonomies Header
– Standard -AIHTaxonomies The characters AIH-HLT-V
– Version Major version – 1 or 2 characters
– Dot-separator The character “.”
– Subversion Minor version – 1 or 2 characters
AIH Data Classes AIH Data Classes are defined by MPAI-TFA Health Data Types
AIH Data Users Different profiles of third-party users can affect the licensing of AIH Data Processing.
– End User Individual who interacts with the AIH platform, primarily via a personal device, providing personal health data and receiving personalised data.
– Non-Profit Entity Entity that is non-profit, e.g., a university.
– Profit Entity Entity that is for profit, e.g., a pharmaceutical company.
– Clinical Entity Entity that looks after the health of patients.
– Authorised Entity Entity that has been authorised by an End User to process some of their AIH Data.
– Caregiver Health providers that interact with the AIH-HSP to provide health and care services to specific End Users (nurses, caregivers, etc.). is folded into 2 intermediaries (back end and 3rd party).
AIH Data Status In terms of Anonymised, Pseudonymised, Identified.
– Anonymised AIH Data may be used if Anonymised.
– Pseudonymised AIH Data may be used if Pseudonymised.
– Identified AIH Data may be used for Identified End User.
AIH Data Usage Types of authorised usage of AIH Data.
– Unrestricted The processed data is open to public or semi-public consultation.
– Pseudonymised The processed data may be published if End User identity are pseudonymised
– Anonymised The processed data may be published if End User identity are anonymised
– Research The processed data may be published if the publication is made on a journal to report research results.
– Patient use The processed data may only be used by the patient or by individual authorised by the patient.
– Health care The processed data may only be used by a Clinical Entity for health-related purpose in the Clinical Entity.
AIH Data Process The types of Processes applied to AIH Data
– ECGProcessingType ECG Data Processing
– EEGProcessingType ECG Data Processing
– GenomicProcessingType Genomic Data Processing
– MedicalImageProcessingType Medical Image Data Processing
DeID&AnonymID ID of DeID&Anonym
DeID&AnonymAlgorithms DeID&Anonym Algorithm
– Data Masking Replaces sensitive data with altered values while preserving the original data structure and format.
– Data Aggregation Combines multiple data records into summary values to reduce individual data exposure.
– Generalisation Substitutes specific data values with broader categories to reduce identifiability.
– Perturbation Introduces controlled modifications or noise into data to prevent accurate inference of original values.
– Tokenisation Replaces sensitive data with surrogate tokens, with original values stored in a secure mapping.
– Hashing Applies a one‑way cryptographic transformation that prevents recovery of the original data.
– Removal of Identifiers Deletes direct identifiers from a dataset to reduce the likelihood of re‑identification.
– K‑Anonymity Ensures each record is indistinguishable from at least k–1 others based on quasi‑identifiers.
– L‑Diversity Ensures that multiple distinct sensitive values exist within each k‑anonymous group.
– Differential Privacy Provides formal guarantees by limiting the influence of any single data subject through calibrated noise.
– Synthetic Data Generation Produces artificial data exhibiting statistical similarity to real data without representing real individuals.
– Homomorphic Encryption Enables computation on encrypted data and returns encrypted results without plaintext exposure.
AIH Data Anomalies Classes of alert messages caused by anomalies in health
  Definition Anomaly examples
– Point Anomaly Individual data points that deviate significantly from the rest of the dataset Sudden spikes in heart rate or blood pressure readings.
– Contextual Anomaly Anomalous data points that in a specific context – may be normal in another. Elevated heart rate during sleep versus during exercise.
– Collective Anomaly A set of related data points that collectively deviate from the expected pattern. A series of abnormal ECG readings indicating a potential cardiac event.
– Medical Condition Anomaly Abnormalities in patient data due to medical conditions. Seizures, falls, arrhythmias, atrial fibrillation, ventricular tachycardia.
– Erroneous Data Anomaly Data errors that may be due to faults or malicious attacks. Anomaly in
– Biorhythm signals (e.g., heartbeat anomalous patterns, respiratory anomalous patterns).
– Multimodal patterns (diverse data sources show conflicting patterns).

<-AI Modules     Go to ToC        Datasets->