AI Transparency on FHIR
0.1.0 - ci-build International flag

AI Transparency on FHIR, published by HL7 International / Electronic Health Records. This guide is not an authorized publication; it is the continuous build for version 0.1.0 built by the FHIR (HL7® FHIR® Standard) CI Build. This version is based on the current content of https://github.com/HL7/aitransparency-ig/ and changes regularly. See the Directory of published versions

ValueSet: Recommended Device.type codes for AI/LLM

Official URL: http://hl7.org/fhir/uv/aitransparency/ValueSet/AIdeviceTypeVS Version: 0.1.0
Standards status: Draft Maturity Level: 0 Computable Name: AIdeviceTypeVS

Subset from HL7, plus those defined here

References

Logical Definition (CLD)

 

Expansion

Expansion performed internally based on codesystem Added Device.type for AI/LLM v0.1.0 (CodeSystem)

This value set contains 18 concepts

LevelSystemCodeDisplay (en)DefinitionJSONXML
1http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS  AI-By-ScenarioClassification by Application Scenario
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Intelligent-Diagnosis-and-TreatmentIntelligent Diagnosis and TreatmentBy analyzing massive volumes of medical data, these AI systems assist doctors in making more accurate diagnostic and treatment decisions.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Medical-Image-AnalysisMedical Image AnalysisLeveraging deep learning technologies, these AI tools automatically identify lesion areas in medical images.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Personalized-TreatmentPersonalized TreatmentThese AI systems create precise patient profiles to formulate personalized treatment plans.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Drug-Discovery-and-DevelopmentDrug Discovery and DevelopmentAI in this category accelerates the screening of candidate drugs and optimizes the design of clinical trials.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Medical-Quality-ControlMedical Quality ControlThese AI tools are used to generate standardized medical document templates and detect defects in medical documents and images.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Patient-ServicesPatient ServicesAI systems here provide patients with services such as intelligent medical guidance, symptom self-assessment, and medical consultation.
1http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS  AI-By-DataTypeClassification by Processed Data Type
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    AI-for-Medical-Imaging-DataAI for Medical Imaging DataIt mainly processes medical imaging data such as X-rays, MRIs, and CT scans.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    AI-for-Physiological-Signal-DataAI for Physiological Signal DataThis type of AI deals with physiological signal data like electrocardiograms (ECG) and electroencephalograms (EEG).
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    AI-for-Medical-Text-DataAI for Medical Text DataIt processes text data such as electronic health records (EHRs) and medical abstracts.
1http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS  AI-By-ModelClassification by Technical Model
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Machine-Learning-ModelsMachine Learning ModelsThey include supervised learning models (e.g., Support Vector Machines (SVM), Random Forests (RF)), which can be used for disease classification and risk prediction; unsupervised learning models (e.g., K-means clustering), which can discover hidden characteristics of patient subgroups; and reinforcement learning models, which can be applied in dynamic treatment plan management.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Deep-Learning-ModelsDeep Learning ModelsExamples include Convolutional Neural Networks (CNNs), which perform excellently in medical image analysis; Recurrent Neural Networks (RNNs) and their variant LSTMs, which are suitable for processing time-series physiological signal data; Generative Adversarial Networks (GANs), which can be used to synthesize training data and alleviate the scarcity of medical data; and Transformer models, which are widely used in multiple tasks such as medical imaging, text analysis, and physiological signal prediction.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Large-Language-ModelsLarge Language ModelsThese models, such as GPT-4 and PaLM, are trained on massive text datasets and can perform various natural language processing tasks, including medical text understanding, generation, and question answering.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Hybrid-ModelsHybrid ModelsThese models combine multiple AI techniques to leverage their respective strengths. For instance, combining CNNs and RNNs can effectively process medical image sequences; integrating machine learning and deep learning models can enhance disease prediction accuracy; and combining rule-based systems with machine learning can improve interpretability and reliability in clinical decision support.
2http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS    Other-AI-ModelsOther AI ModelsThis category includes other AI models not covered above, such as graph neural networks (GNNs) for modeling complex relationships in medical data, and evolutionary algorithms for optimizing treatment plans.
1http://hl7.org/fhir/uv/aitransparency/CodeSystem/AIdeviceTypeCS  Artificial-IntelligenceAll kinds of Artificial Intelligence

Explanation of the columns that may appear on this page:

Level A few code lists that FHIR defines are hierarchical - each code is assigned a level. In this scheme, some codes are under other codes, and imply that the code they are under also applies
System The source of the definition of the code (when the value set draws in codes defined elsewhere)
Code The code (used as the code in the resource instance)
Display The display (used in the display element of a Coding). If there is no display, implementers should not simply display the code, but map the concept into their application
Definition An explanation of the meaning of the concept
Comments Additional notes about how to use the code