EU AI ACT · EU MDR / IVDR · FDA · INTERNATIONAL
Artificial Intelligence in Medical Devices
For a conventional device you demonstrate safety by showing how it works. For a device built on a learned model you cannot: nobody wrote the rules it applies. This course is about what you demonstrate instead — where the data came from, how the truth was established, how performance was measured and on whom, what a human can still catch, and how you would know if it stopped working.
Scope
What the course covers
Where a requirement is quoted, it is quoted in full and then explained. Where something is an interpretation rather than a rule, the course says so.
01
Qualification and classification
Whether the product is a device and an AI system, the MDR class and the AI Act risk tier, and how the two scales meet at the notified body.
02
The timeline, in one place
Every date and every document status lives in a single module, so the rest of the course stays valid as the field moves.
03
Data governance
Provenance, legal basis, representativeness, annotation and ground truth, pre-processing, and the splits that decide whether your performance figure means anything.
04
Bias and evaluation
Bias examined rather than declared absent, the metrics that support a claim, calibration, prevalence, subgroups, and the analysis plan written before the test set is opened.
05
Risk, robustness and security
Hazards that arise from correct operation, human oversight as a control rather than a sentence, and the attacks that exist only because there is a model.
06
Clinical evidence and change control
Study designs that work for AI, why equivalence rarely does, and how to update a model without going back through conformity assessment.
07
Generative systems
Where the standard toolkit stops applying, and what can be taken to a notified body today.
08
The other markets
The United States route, the international layer, and the markets that accept somebody else’s authorisation instead of their own.
Outline
16 modules, in the order the process runs
The course follows the sequence of the work rather than the numbering of the regulation, and ends where the work ends.
Introduction
Four ways an AI submission comes backQualification and classification
Device, AI system, class and risk tierThe regulatory landscape and the clock
What applies, and from whenIntended purpose and requirements
The most expensive sentence in the fileData governance
Sources, annotation, pre-processing, splitsBias, model development and evaluation
Where the numbers come fromQMS and software lifecycle for AI
What changes in what you already haveRisk management, robustness and cybersecurity
When the failure mode is statisticalTransparency, usability and human oversight
Getting the information to the person who actsClinical evaluation and performance evidence
From a correct output to a patient benefitChange control and PCCP
Updating a model without starting againPost-market monitoring and vigilance
Watching a device that can become wrongGenerative AI, foundation models and LLMs
Where the toolkit stops reachingThe FDA route
The same evidence, a different containerOther markets and the international layer
One file, several submissionsTechnical documentation and the notified body
Assembling one file, and defending itFinal test
Covering the whole courseOutcomes
What you will be able to do
01
Decide what you are
Qualify and classify an AI-enabled device under both regulations, and write the determination so it survives an audit.
02
Write an intended purpose that holds
Eight elements, including the two that are almost always missing: the level of autonomy and the input constraints.
03
Document data the way a reviewer reads it
Provenance, representativeness demonstrated rather than asserted, and splits that are independent in the way that matters.
04
Defend a performance figure
Know which metric supports which claim, why predictive values do not travel between populations, and what a pre-specified analysis plan protects you from.
05
Make oversight a control
Build the four conditions that turn human oversight from a sentence in the manual into something that works at the moment of use.
06
Plan the updates
Write a change control plan that gets accepted, and know what can never go inside one.
07
Detect a device becoming wrong
Build monitoring that finds degradation when nothing has failed and no complaint has arrived.
08
Take the same file to several markets
Build a jurisdiction-neutral core, and understand why the order in which you approach markets is a strategic decision.
Format
Designed to be worked through, not watched
Self-paced and fully online, with progress saved to your account and every module replayable. Transcripts are downloadable, so the content stays searchable long after the video is finished. Access does not expire, and revisions made after your purchase are included.
Audience
Who the course is for
- Regulatory affairs professionals working on software or AI-enabled devices
- Quality managers extending an existing system to cover AI
- PRRCs responsible for devices that use machine learning
- Clinical and medical affairs staff building the evidence for an AI claim
- Engineering and data science leads who have to produce the documentation
- Consultants and notified body personnel who assess these files
It assumes familiarity with medical device regulation and with a quality management system. It does not assume prior study of the articles and annexes it covers.
Certificate
Verifiable by a third party
On passing the final test, the certificate carries a verification code derived from your name and the date of completion.
An employer, an auditor or a client can confirm it on a public page, and it can be filed as training evidence under your quality system.
MDR-XXXX-XXXX · mdregulatory.com/verify
Questions
Before you enrol
Do I need to understand machine learning?
No. The course explains the technology only where a regulatory consequence follows from it, and assumes no background in it.
The AI Act keeps moving. Will this course date?
Every date and document status sits in one module, deliberately. When something changes, that module is updated and the rest of the course remains valid.
Does it cover devices outside Europe?
Yes. Three modules cover the United States route, the international layer, and the markets that rely on an authorisation obtained elsewhere.
Does it cover generative AI and large language models?
Yes, in a module of its own — including what can realistically be brought to a notified body today, and what cannot.
We already have a certified quality system. How much changes?
Less than most people expect. The course maps the new obligations onto the procedures you already run, and identifies the few that are genuinely additional.
How long do I have access?
Access does not expire. The course stays in your account, including any updates made to the modules after your purchase.
Is there a certificate?
Yes. Passing the final test issues a certificate with a verification code, which anyone can check on our verification page.
Can I watch the modules in any order?
The modules unlock in sequence, because each one builds on the previous. Once unlocked, a module stays available and can be rewatched.
Do I need any prior knowledge?
The course assumes you work in the sector and know what a technical file and a notified body are. No knowledge of the specific technology is assumed.
Can my company buy several seats?
Yes. Write to us and we will set up team access with a single invoice.
Course and kit
The course explains the reasoning. The kit gives you the structure.
The workbook included with the course is a gap analysis of your own documentation, arranged in the order an assessment runs. Working through it produces a list of what is missing or partial in your file.
That list maps to the templates in the AI Act Technical Documentation Kit, MDR Technical File Kit.
Enrol
Everything an AI-enabled device needs, in the order a reviewer reads it.
16 modules, 5 hours 20 minutes of video, downloadable transcripts, workbook and certificate. Self-paced, with no expiry.