Updated:
Medical AI glossary
Medical AI terms you meet in proposals and meetings, collected to the extent needed for a practical assessment. You can jump straight to a term by its name.
Terms tied to regulation are given as of the time of writing, based on the Singaporean documents read then. They are summaries, not the text of the law. Before you decide, check the primary source and, where needed, ask a specialist.
Regulation and certification
- HSA (Health Sciences Authority)
-
The authority whose guidelines on software medical devices, risk classification and change management this site cites (GL-04, GL-07 and GN-37). It also receives reports of adverse events from the use of medical devices, as the Ministry of Health’s AI guidelines describe.
Why it mattersIts guidance documents page is where to read the current text before you rely on a summary, including this one.
Source: hsa.gov.sg (Opens in a new tab)
- SaMD (Software as a Medical Device)
-
HSA’s GL-07 defines a standalone medical mobile application as software intended for one or more medical purposes that works by itself and is not meant to control or affect the operation of other hardware medical devices, and notes these are commonly called SaMD. It runs on general computing devices such as laptops, tablets and desktops.
Why it mattersAsk a vendor whether its product is SaMD, and on what intended use it bases the answer.
- AI-enabled and ML-enabled medical device
-
GL-04-R4 defines an artificial intelligence-enabled medical device (AIMD) as a medical device that uses AI technology to achieve its intended medical purpose, and a machine learning-enabled medical device (MLMD) as one that uses machine learning, in part or in whole, to achieve its intended medical purpose.
Why it mattersThese are the products to which the extra ML requirements in GL-04 section 9 apply, such as documentation of the model and monitoring after deployment.
- Intended use
-
As defined in the Health Products (Medical Devices) Regulations 2010 and quoted in GL-07: the objective intended use or purpose of a medical device, as reflected in the specifications, instructions and information provided by the product owner.
Why it mattersWhether software is a medical device turns on its intended use, so compare the vendor’s written intended use with what your staff will actually do with it.
- Risk class (A to D)
-
GL-07-R2 sets the class of a SaMD from the significance of its information (treat or diagnose, drive, or inform clinical management) and the state of the patient’s situation (critical, serious or non-serious). GL-04-R4 says Class B, C and D software medical devices require product registration.
Why it mattersThe class tells you how closely a product is regulated, so ask for it with the reasoning.
Source: isomer-user-content.by.gov.sg (Opens in a new tab) / isomer-user-content.by.gov.sg (Opens in a new tab)
- Clinical decision support software (CDSS)
-
Software that supports clinical decisions. GL-07 provides clarity on when CDSS qualifies as a regulated medical device and when it does not.
Why it mattersA vendor’s statement that its tool is “only decision support” is a claim to test against GL-07, not a reason to skip the device question.
- Change notification
-
HSA’s process for changes to a registered device. GL-04-R4 says a change notification is required for any change to a registered ML-enabled device, and the category of change (technical, review or notification) follows GN-21.
Why it mattersAsk the vendor how it handles model updates before one reaches your wards.
- Change Management Program (CMP)
-
An optional HSA pathway for SaMD, including machine-learning enabled SaMD, introduced in GN-37-R1 (December 2024). It lets a manufacturer have anticipated “pre-specified changes” approved in advance. The guidance excludes continuous learning and generative AI.
Why it mattersIf a vendor says its updates are pre-approved, ask whether it means a CMP and which changes are covered.
- AIHGle 2.0
-
The Artificial Intelligence in Healthcare Guidelines, version 2.0, published by the Ministry of Health and HSA in March 2026. It sets out responsibilities for developers, deployers and users and adds chapters on continuous-learning, generative and direct-to-consumer AI.
Why it mattersIt is guidance, not a statute, but it is the clearest statement of what Singapore’s health authorities expect of a hospital that deploys AI.
Source: isomer-user-content.by.gov.sg (Opens in a new tab) / moh.gov.sg (Opens in a new tab)
- AI-MD Exemption Sandbox
-
A sandbox launched by MOH and HSA in February 2026, according to AIHGle 2.0, that lets selected public healthcare entities deploy low-to-moderately-low-risk AI medical devices across multiple institutions without full licensing and registration requirements.
Why it mattersIt does not mean AI medical devices are generally exempt. It is for selected public healthcare entities.
Data and privacy
- PDPC (Personal Data Protection Commission)
-
The commission that publishes the Guide on Managing and Notifying Data Breaches under the PDPA cited on this site, and to which notifiable breaches are reported. AIHGle 2.0 names it as the authority for data-related incidents.
Why it mattersIts guide sets the clock for breach assessment and notification, so read the current revision.
Source: pdpc.gov.sg (Opens in a new tab)
- Notifiable data breach
-
According to the PDPC guide (revised 15 March 2021), an organisation must assess within 30 calendar days of having credible grounds to believe a breach occurred whether it is notifiable, and notify the PDPC within 3 calendar days after determining that it is. A breach is notifiable when it is likely to cause significant harm to affected individuals or is of significant scale, meaning the personal data of 500 or more individuals.
Why it mattersIt tells you what to prepare in advance: a named assessor, documented steps and a way to file within days.
Source: pdpc.gov.sg (Opens in a new tab)
Standards and terminology
- Deployer
-
In AIHGle 2.0, the healthcare organisation that deploys an AI solution, as distinct from the developer (for example a manufacturer) and the user (a healthcare professional).
Why it mattersMost readers of this site are deployers. The guidelines give deployers duties on governance, testing, training, monitoring and incident response.
- Human-in-the-loop, over-the-loop, out-of-the-loop
-
AIHGle 2.0 Table 4 distinguishes three levels of human oversight: in the loop (the human decides and the AI only recommends), over the loop (the human monitors and can take over) and out of the loop (no human oversight of execution). For clinical use it says deployers should not deploy AI that makes clinical decisions independently without a human making the final decision.
Why it mattersIt is a simple way to describe, in a proposal, who really makes the decision.
- AI registry
-
AIHGle 2.0 says a hospital’s governance platform should maintain a registry of all AI solutions deployed in the organisation, including each solution’s risk classification.
Why it mattersYou cannot monitor tools you do not know about. The registry is the first list to create.
- Hallucination
-
AIHGle 2.0 uses the word for generative AI models that generate and present incomplete or fictitious content as facts, where users may not notice the inaccuracies.
Why it mattersIt is why AI drafts need a fact-checking step before they are used or entered into a record.
- Model drift
-
AIHGle 2.0 lists model drift among the performance indicators a hospital’s monitoring process should track after deployment.
Why it mattersAsk the vendor how drift is detected and who tells you when performance changes.
- ISO 13485 quality management system
-
GL-04-R4 says all activities for designing, developing, training, validating, retraining and deploying a machine-learning-enabled medical device must be managed under an ISO 13485 based quality management system.
Why it mattersIt is a reasonable thing to ask a vendor about: whether its development of the product follows such a system.
This glossary is not a translation of the Korean or Japanese edition, and not a copy of the U.S. or India edition. Each regulatory entry is written from a Singaporean official document that was opened and read. Anything we could not confirm is left out.