Skip to main content

AI/ML-Based Medical Devices

Korea was among the first regulators globally to issue dedicated AI/ML medical device guidance (initial guidance 2017; updated 2021, 2024).

Classification​

Same SaMD grade criteria, with attention to:

  • Whether AI output is used directly for clinical decisions (higher risk)
  • Locked (static) vs adaptive (changing) algorithm
  • Clinical consequence of an incorrect AI output

Key application requirements for 품목허가​

  • Algorithm description and validation data
  • Training data characteristics and representativeness
  • Test dataset performance metrics
  • Explainability / transparency documentation
  • Post-market performance monitoring plan

Post-market monitoring​

MFDS requires ongoing monitoring of AI algorithm performance after approval, including: • Real-world performance monitoring against defined performance thresholds • Drift detection — monitoring for degradation in algorithm accuracy over time • Planned re-validation when significant data shifts occur or when performance thresholds are approached • Documentation and reporting of monitoring results to MFDS as required • Manufacturer must commit to algorithm updates or corrective actions if performance degrades below acceptable limits

MFDS issued dedicated guidance documents: • Initial AI/ML Medical Device Guidance (2017) • Updated guidance (2021) • Most recent update (2024)

Manufacturers should consult the latest version on the MFDS website for current requirements.

For the latest AI/ML guidance requirements, manufacturers should: • Consult the MFDS website directly for current guidance versions • Review any device-specific annexes or technical guidance released after 2024 • Engage with MFDS via pre-submission consultation to clarify AI/ML-specific requirements for their device type

Note: The regulatory landscape for AI/ML devices continues to evolve; early engagement with MFDS is strongly recommended.