Skip to content
Back to Blog
Bio
2026-08-318 min read1

New Approval Guideline for Professional Digital Medical Device Software: What SME SaMD and Medical AI Firms Must Prepare Under the Digital Medical Products Act

A practical breakdown of the new approval and review guideline for professional-use digital medical device software under Korea's Digital Medical Products Act, and what SME SaMD and medical AI firms should prepare — from evidence design and quality systems to certification cost support programs.

KITIM Consulting Team

What Changed After the Digital Medical Products Act

With the Digital Medical Products Act in force, software-based medical products are no longer treated as just one branch under the Medical Devices Act. The law defines 'digital medical devices' and, within that, 'digital medical device software' as distinct categories. Now that the enforcement decree and enforcement rules are both in place, the discussion has moved past interpreting the text — these rules are being applied in actual approval reviews.

Two things changed in substance. First, review criteria now assume the traits specific to software: frequent updates and performance that shifts with learning. Second, the range and depth of required submission materials rose accordingly. This is exactly why experience with hardware device approvals alone no longer carries a company through.

The New 'Professional Use' Category

The guideline established in 2026 treats professional-use digital medical device software as its own category. The dividing line is not simply who operates the product, but how deeply it intervenes in clinical judgment. Consumer-facing software that supports lifestyle management or offers reference information sits at one end; professional-use software that directly informs a clinician's diagnostic or treatment decision faces a materially higher evidence bar.

When determining whether your product falls into this category, check the following:

  • Are end users limited to healthcare professionals, or do patients interpret the output directly?
  • Does the output map directly onto a decision — a diagnosis, a lesion location, a severity grade?
  • Can the clinician independently review the basis for the software's conclusion?
  • How severe is the potential patient harm if the software fails?
  • The third point deserves particular attention. Assistive reading tools that a clinician can verify against the original image or source data tend to be assessed at lower risk. Software that outputs a conclusion without exposing its basis pushes both the classification and the documentation burden upward.

    What the Review Dossier Now Requires

    The required materials fall into three groups.

  • Performance evaluation: Sensitivity, specificity, and AUC reported with confidence intervals, validated on an independent dataset that was not used in training. Single-site data alone frequently draws a finding that generalizability has not been demonstrated.
  • Usability engineering: Formative and summative evaluation under IEC 62366-1, with validation records covering representative user groups.
  • AI-specific items: The provenance and legal basis of training data, labeling procedures and labeler qualifications, bias management across age, sex, equipment manufacturer, and institution, plus an algorithm change management plan.
  • The guideline finalized after the August public consultation applies, in principle, to new applications. It is not retroactively imposed on already-approved products as a batch — but the new standard takes effect at the point of a change approval or renewal, so the practical grace period is shorter than it looks.

    Where SMEs Actually Get Stuck

    First, the boundary around retrospective data. Assistive reading software that automates an already-established interpretation standard can often demonstrate effectiveness using multi-site retrospective data. Products that propose a novel predictive marker, or that change the treatment plan itself, are far more likely to be asked for prospective confirmatory evidence. Making this determination late in development means redesigning the clinical strategy from scratch.

    Second, change approval overload. Filing a change approval for every performance improvement can tie up regulatory resources three or four times a year. Defining the scope of permitted changes, the verification method, and the minimum acceptable performance in advance lets a substantial share be handled through internal controls.

    Third, an underbuilt software quality system. Configuration management, traceability, and cybersecurity documentation left at the level of ordinary development practice are a recurring source of GMP findings.

    The classic failure involves training data. We regularly see companies spend six months or more recollecting data because what they gathered under a research-only IRB approval could not be used as approval evidence.

    A Preparation Roadmap and Available Support

    A realistic timeline runs six to twelve months.

  • Months 1–2: Confirm product classification and professional-use status; fix the regulatory pathway
  • Months 2–5: Design the clinical evidence strategy; settle data sourcing and consent scope
  • Months 4–9: Build the GMP and software quality system; conduct usability evaluation
  • Months 9–12: Compile the technical file, submit, and respond to deficiencies
  • Using the MFDS pre-submission consultation early in development lets you confirm classification and the expected evidence level before committing engineering resources, which sharply reduces redesign risk. Pairing the effort with government programs that subsidize medical device certification and approval costs can offset a meaningful share of consulting and testing expenses, so it is worth aligning the approval timeline with the support program calendar.

    ---

    KITIM supports SaMD and medical AI companies across the full approval pathway — from product classification and clinical evidence design to GMP and software quality system implementation and linkage with certification cost support programs. Fixing the regulatory pathway during development is the cheapest moment to do it. If you have a product under review, please get in touch and we will walk through the right order of preparation with you.

    Digital Medical Products ActDigital Medical Device SoftwareMedical AISaMD ApprovalMFDS Guideline
    매일 자동 업데이트

    이 분야 정부지원사업, AI가 찾아드립니다

    3분 기업진단만 완료하면 귀사에 맞는 공고를 적합도 점수와 함께 추천합니다. 무료입니다.

    AI 맞춤 공고 무료로 받기

    Need Consulting?

    Our technology innovation consultants will propose the optimal solution for your company.