How a Korean Digital Health Startup Cleared MFDS SaMD Pathway in 5 Months
A Korean AI radiology startup achieved MFDS Class III SaMD approval in 5 months — faster than the typical 8–14 month timeline. Here's how the strategy worked.
Background
The startup developed an AI-assisted detection software for chest X-ray nodule identification. The device is Class III (highest risk) due to direct clinical decision support implications.
Challenges:
- Class III SaMD typically takes 8–14 months in Korea
- AI/ML algorithms require change management framework
- Clinical evidence needed Korean population validation
- Startup constraints: limited cash runway, need fast revenue
Strategy: 3 Parallel Workstreams
Instead of sequential Phase 1 → Phase 2 → Phase 3, the team ran 3 parallel workstreams:
Workstream 1: Algorithm Validation (Months 0–3)
- Korean validation dataset assembly (5,000 chest X-rays from 4 Korean hospitals)
- IRB approval at lead clinical site
- Algorithm performance validation per MFDS guidance
- Bias analysis across Korean demographic subgroups
Workstream 2: QMS and KGMP (Months 0–4)
- ISO 13485 QMS implementation (no prior QMS — startup)
- Korean SOPs developed in parallel (15 SOPs)
- KGMP gap analysis
- MFDS KGMP inspection (Month 4)
Workstream 3: Documentation and Submission (Months 0–4)
- Korean STED preparation
- Clinical evaluation report with Korean validation data
- Cybersecurity submission package (preparing for 2026 requirements)
- Algorithm change management plan (PCCP-equivalent)
Critical Decisions That Saved Time
Decision 1: Parallel KGMP from Day 1
Most startups wait until product registration is filed before scheduling KGMP. The team scheduled KGMP inspection in Month 1 for Month 4 — running parallel to documentation prep.
Time saved: 4 months.
Decision 2: Korean Validation Study Pre-Submission
Many startups complete validation post-submission, then face deficiency notices requesting additional data.
The team conducted Korean validation before submission, including:
- 5,000 image dataset with Korean demographic representation
- Subgroup analysis (age, gender, comorbidity)
- Bias assessment with fairness metrics
- Performance comparison with radiologist baseline
Time saved: 60–90 days (avoided 1–2 deficiency notice rounds).
Decision 3: Algorithm Change Management Plan in Initial Submission
MFDS announced PCCP-equivalent framework in 2025. The team included algorithm change management plan in initial submission, anticipating post-market update requirements.
Time saved: Avoided needing to file new submissions for planned algorithm updates.
Decision 4: Pre-Submission MFDS Consultation
Engaged MFDS pre-submission consultation in Month 2 to clarify:
- SaMD classification (Class III confirmed)
- Clinical evidence expectations
- Algorithm change management acceptable structure
- Cybersecurity submission scope
Time saved: Clarified expectations early, avoided major rework.
Timeline
| Month | Workstream Activities |
|---|---|
| Month 0 | Project kickoff, gap analysis, dataset assembly initiated |
| Month 1 | KGMP application filed, QMS implementation started, IRB approval received |
| Month 2 | MFDS pre-submission consultation, validation study enrollment |
| Month 3 | Validation study complete, Korean STED first draft, QMS mock audit |
| Month 4 | KGMP inspection passed, submission filed |
| Month 5 | MFDS clearance — KMD number issued |
Cost
| Cost Category | Amount |
|---|---|
| Leanabl consulting (end-to-end) | $120K |
| Korean validation dataset | $35K |
| KGMP inspection fees + travel | $18K |
| MFDS official fees (Class III) | $12K |
| Translation | $20K |
| Total | $205K |
For comparison: typical Class III SaMD program runs $250K–$400K and 10–18 months. The team achieved approval at 18% lower cost and 60% faster.
Results
Regulatory Outcomes
- ✅ MFDS Class III SaMD approval in 5 months
- ✅ Zero deficiency notices on first submission
- ✅ PCCP-equivalent algorithm change management framework approved
- ✅ KGMP certificate valid for 3 years
Business Outcomes
- ✅ Korean market entry 6+ months ahead of competitors
- ✅ Series A funding accelerated (regulatory de-risking)
- ✅ HIRA reimbursement application initiated (Month 6)
- ✅ Initial hospital deployments at lead clinical sites
Lessons Learned
Lesson 1: Parallel Workstreams Beat Sequential
The standard sequential approach (QMS → Validation → Documentation → Submission) takes 12+ months. Parallel execution requires more upfront coordination but cuts 4–6 months.
Lesson 2: Korean Validation Before Submission
MFDS expects Korean-population validation for AI/ML devices. Pre-submission validation avoids 60–90 day deficiency notice cycles.
Lesson 3: PCCP-Equivalent Framework Now
Even if FDA-style PCCP isn't strictly required yet by MFDS, including an algorithm change management plan signals regulatory sophistication and prepares for inevitable updates.
Lesson 4: Pre-Submission Consultation Worth It
MFDS pre-submission consultation costs nothing but time. Clarifying expectations early prevents costly rework.
Lesson 5: Startup Speed Requires Senior Practitioners
The startup engaged senior regulatory practitioners (not junior teams) from Day 1. The premium ($120K vs $80K for junior-led program) recovered through faster timeline and zero deficiency notices.
What Worked Best
- Korean validation in lead clinical sites — generated post-market commercial relationships
- MFDS pre-submission consultation — set expectations with regulatory team
- Algorithm change management plan — future-proofed for updates
- Parallel workstreams — required disciplined PM, recovered 4–6 months
- Senior consultants — premium investment justified by outcomes
What Could Have Gone Better
- Korean validation dataset cost higher than anticipated (~30% over budget)
- Cybersecurity submission scope expanded mid-project (anticipating 2026 requirements)
- KGMP inspection scheduling tighter than ideal — recommend 12+ week lead time
Frequently Asked Questions
Q: Is this timeline replicable for other AI/ML devices?
A: For Class III SaMD with Korean validation data in hand: yes, achievable. For devices needing Korean clinical study from scratch: 8–12 months more realistic.
Q: What was the most critical decision?
A: Parallel KGMP from Day 1. Sequential approach would have added 4 months minimum.
Q: Did the startup need a Korean entity?
A: As a Korean startup, they had Korean entity already. For foreign manufacturers, KLH (Korea License Holder) serves equivalent function — independent KLH recommended for AI/ML devices given complexity of post-market obligations.
Q: How does HIRA reimbursement timeline overlap?
A: HIRA submission initiated post-MFDS approval. HIRA review typically 6–18 months, depending on health economics evidence and benchmark device pricing.
Q: Can Leanabl support similar AI/ML SaMD projects?
A: Yes. Leanabl's Korea SaMD Approval solution is designed specifically for AI/ML medical device Korea entry.
How Leanabl Helps
- Korea SaMD Approval — AI/ML medical device pathway
- Korea Medical Device Registration — full Class III/IV programs
- Regulatory Pathway Strategy — initial strategy
Contact Leanabl for AI/ML SaMD scoping.
Last updated: 2026-05-15.
Have a regulatory question?
Talk to a Korea regulatory specialist about your device, your timeline, or your next submission.
Talk to a specialist
