Short Answer
The class of a medical-AI software is determined not by its code but by its intended use statement. Under EU-MDR Annex VIII Rule 11, software that provides information for diagnostic or therapeutic decisions is by default Class IIa; however, if the impact of those decisions can cause a serious deterioration in health or a surgical intervention, it becomes Class IIb, and if it can cause death or irreversible deterioration, it becomes Class III. Software that serves only an administrative or simple reference function remains Class I. In practice, the fault line between IIa and IIb runs right through the middle of most surgical-planning and decision-support software: what is decisive is whether the output is considered "information that shapes a surgical intervention." For this reason, classification is not a technical labeling exercise but a strategic decision managed by writing the intended use statement narrowly and defensibly.
Serteser Danismanlik is run by a biomedical engineer (BME MSc) who has developed a medical-AI medical device and published it in a peer-reviewed international journal. We build the intended use statement of your SaMD/AI devices, the classification rationale (Rule 11 / MDCG 2019-11), and the clinical-validation evidence to support it in a form that is defensible in the regulatory file.
Classification is one of the most expensive late decisions made by early-stage medical-AI teams. The class determines the depth of clinical evidence the product will require, the scope of notified body scrutiny, and therefore the entire regulatory budget and timeline. Yet most teams make this decision while preparing the file, that is, at the most expensive moment.
This article clarifies how classification works and why it is less a technical label than an intended-use writing decision.
The Class Is Determined by Intended Use, Not Code
The intuitive misconception in classification is to assume that the class is determined by how "advanced" the software is. What is decisive is not the technology but the declared intended use: for whom, for which clinical decision, and at what level of autonomy does the software provide support? The same segmentation model can fall into a low class when defined as a "preliminary visualization reference" and into a high class when defined as a "measurement that determines the surgical resection plan."
For this reason, the first input to classification is not statistics or architecture but a single sentence: the intended use statement. This sentence determines both the class and the scope of the validation study you will later design.
Rule 11: The Software-Specific Classification Rule
Rule 11 in EU-MDR Annex VIII divides decision-support software into four tiers:
- Class I: Software that serves only an administrative, communication, or simple reference function and does not feed a diagnostic/therapeutic decision.
- Class IIa: Software that provides information for diagnostic or therapeutic decisions. This is the baseline class of most decision-support and planning software.
- Class IIb: If the impact of those decisions can cause a serious deterioration in health or a surgical intervention.
- Class III: If the impact of those decisions can cause death or an irreversible deterioration.
In Turkey, this framework is not a separate logic: the TITCK Medical Device Regulation transposes EU-MDR one-to-one, so the same Rule 11 and the same MDCG 2019-11 (software qualification and classification guidance) logic apply. This means the classification work is also reusable for EU CE.
IIa or IIb: Where Is the Fault Line?
In practice, the debate is almost always between IIa and IIb, and it rests on a single phrase: surgical intervention. If the software's output is "supporting information" that feeds a decision the surgeon makes and verifies independently, IIa is defensible. If the output is positioned as information that directly shapes a surgical intervention (such as tunnel position, resection plane, or implant placement), a reviewer may request IIb.
The critical point: this difference lies not in the technology but in the language. What an intended use statement that aims to preserve IIa must do:
- Define the output as "supporting/reference information," not as "determining the plan."
- Make the surgeon's independent review and verification (human-in-the-loop) mandatory.
- Explicitly exclude autonomous decision-making and intraoperative control from scope.
- Declare that final responsibility rests with the surgeon.
An honest note: if the planning outputs genuinely shape the surgical intervention, a IIb outcome is not devastating. IIa and IIb require the same process skeleton (ISO 13485 + notified body + technical file); the difference is in the threshold of expected clinical evidence. That is why the goal is to write the class narrowly and defensibly first, then confirm it in a TITCK pre-consultation.
The Intended Use Statement: The Highest-Leverage Decision
The way to manage classification is not to change the product but to write the intended use with discipline. This sentence should not be translated into marketing language before it is frozen, because marketing must derive from the intended use, not the other way around. In practice, the most common mistake is that the label or website copy attributes a role to the software beyond its intended use ("plans the surgery," "makes a diagnosis") and this language unintentionally raises the class.
To build your SaMD device's intended use statement in a way that preserves the class and aligns with your clinical-validation study, you can request a free 15-minute scoping.
Decision Table: Which Class Are You Closest To?
| If your intended use output is... | Likely class |
|---|---|
| Only administrative / archival / simple visualization | Class I |
| Information feeding a diagnostic or therapeutic decision, surgeon verifies independently | Class IIa |
| Shapes a decision that could cause serious deterioration or a surgical intervention | Class IIb |
| Shapes a decision that could cause death or irreversible deterioration | Class III |
This table is not a diagnosis but a starting point: the actual class depends on the exact wording of the intended use statement and on the TITCK/notified body assessment.
Conclusion: Classification Is a Writing Discipline
The class is the result not of how advanced your product is but of what your intended use claims. Writing this sentence early, narrowly, and aligned with your clinical-validation plan is free but is the highest-leverage regulatory decision; writing it late and loosely means a difference of tens of thousands of dollars and months of delay down the line. The classification rationale is the first page of your file, and it cannot easily be reversed afterward.