Classify a diagnostic AI tool under software as a medical device risk categories to determine its regulatory route.
Healthcare & Pharmaceutical Management
Diagnostic AI Validation and Regulatory Approval Pathways
Take a diagnostic AI tool from analytical validation through clinical study design to FDA and EU regulatory submission, with methods for bias testing and post-market drift monitoring.
🗓️Duration5 training days
📚Content4 modules · 8 sessions
🎓On completionAccredited attendance certificate
About the programme
Course Overview
A diagnostic algorithm that performs well on a training dataset can still fail regulatory review or, worse, fail silently on a patient subgroup it was never properly tested against. This course equips regulatory affairs specialists, clinical scientists and product teams to validate diagnostic AI tools rigorously and carry them through the appropriate approval pathway. Participants learn to classify a candidate device under software as a medical device risk categories and compare how the FDA and the EU's Medical Device and In Vitro Diagnostic Regulations treat algorithmic tools differently. The course covers validation study design, including how to select a reference standard and calculate sensitivity, specificity and predictive values across representative populations. A dedicated module addresses algorithmic bias and dataset shift, teaching subgroup performance analysis so accuracy claims hold across age, sex and ethnicity strata rather than only on average. Participants also learn to draft a predetermined change control plan for algorithms that continue learning after deployment, and to build explainability outputs that calibrate clinician trust appropriately. The course closes with post-market monitoring: detecting performance drift in real-world data and maintaining the regulatory submission as the algorithm or its environment changes.
Expected Learning Outcomes
Compare FDA and EU MDR/IVDR submission requirements for a specific diagnostic algorithm.
Design an analytical and clinical validation study with an appropriate reference standard and population sample.
Conduct a subgroup performance analysis to detect algorithmic bias across demographic strata.
Draft a predetermined change control plan governing how an adaptive algorithm may evolve post-approval.
Build explainability outputs that calibrate clinician trust without overstating the algorithm's certainty.
Design a post-market monitoring plan that detects performance drift and triggers a regulatory review.
Who Should Attend
Regulatory affairs specialists submitting diagnostic AI devices for approval
Clinical scientists designing validation studies for algorithmic diagnostic tools
Health-tech product managers bringing AI-enabled diagnostics to market
Radiology and pathology informatics leads evaluating AI tools for clinical deployment
Quality assurance staff responsible for post-market surveillance of software medical devices
Data scientists translating model performance metrics into regulatory evidence
Course Modules
Select any module to see its sessions and points.
01Diagnostic AI Fundamentals and Regulatory Classification
2 sessions · 8 points
Session 1Software as a Medical Device Risk Categorisation
- Apply the international risk categorisation framework to classify a diagnostic algorithm's clinical significance.
- Distinguish tools that inform a clinical decision from those that drive it autonomously for classification purposes.
- Document the intended use statement that anchors a device's risk category and regulatory scope.
- Identify how a change in intended use or population can shift a device into a higher risk class.
Session 2Regulatory Pathways: FDA, EU MDR/IVDR Comparison
- Compare the FDA's premarket clearance and approval routes available to a diagnostic AI device.
- Map EU Medical Device and In Vitro Diagnostic Regulation conformity assessment routes for AI software.
- Identify documentation common to both jurisdictions to avoid duplicating regulatory evidence generation.
- Plan a submission timeline that accounts for notified body or regulator review cycles.
02Clinical Validation Study Design
2 sessions · 8 points
Session 1Analytical and Clinical Validation Study Design
- Select a reference standard appropriate to the diagnostic claim being validated.
- Calculate sample size needed to demonstrate sensitivity and specificity with adequate statistical power.
- Design a study population that reflects the real-world case mix the device will encounter in use.
- Distinguish analytical validation of algorithm output from clinical validation of patient outcomes.
Session 2Bias, Dataset Shift and Subgroup Performance Analysis
- Test algorithm performance separately across age, sex, ethnicity and disease severity subgroups.
- Detect dataset shift between training data and the deployment population before go-live.
- Document identified performance gaps and the mitigation or labelling response chosen for each.
- Design ongoing subgroup monitoring to catch bias that emerges only after wider deployment.
03Predetermined Change Control and Explainability
2 sessions · 8 points
Session 1Predetermined Change Control Plans for Adaptive Algorithms
- Draft a predetermined change control plan defining the boundaries within which an algorithm may adapt.
- Specify the retraining, validation and documentation steps required before each permitted update.
- Define triggers that require a fresh regulatory submission rather than a permitted change.
- Align change control governance with internal quality management system change control procedures.
Session 2Explainability and Clinician Trust Calibration
- Select an explainability method appropriate to the algorithm type and clinical use case.
- Design output displays that communicate confidence level without implying false certainty.
- Test clinician interpretation of explainability outputs to confirm they support rather than mislead judgement.
- Document explainability limitations transparently within labelling and user instructions.
04Post-Market Surveillance and Real-World Monitoring
2 sessions · 8 points
Session 1Post-Market Performance Monitoring and Drift Detection
- Design a real-world performance monitoring plan that tracks accuracy against a clinical outcome benchmark.
- Set statistical thresholds that flag performance drift requiring investigation or intervention.
- Investigate a flagged drift event to determine whether it stems from data, population or model change.
- Decide when drift requires model retraining, temporary suspension or a labelling update.
Session 2Incident Reporting and Regulatory Submission Maintenance
- Establish an adverse event and malfunction reporting process meeting regulator timelines.
- Maintain the technical file and clinical evidence dossier as the algorithm evolves under its change plan.
- Coordinate periodic safety update reporting with regulatory affairs and quality teams.
- Plan a device withdrawal or suspension protocol for use if post-market data reveals unacceptable risk.
What the participant receives
4 course modules
A structured syllabus
8 training sessions
across 5 days
32 detailed points
Applied, detailed content
Accredited attendance certificate
On completing the programme
Complete your registration
We will contact you within one business day to confirm.
Ready to start?
Reserve your seat and start building the skill.