Overview
CEN ISO/TS 24971-2:2026 provides essential guidance for applying the ISO 14971 risk management process specifically to medical devices that incorporate machine learning in artificial intelligence (MLMD). As AI and machine learning continue to transform healthcare, understanding and managing their unique risks is critical for both manufacturers and users. This technical specification helps stakeholders navigate the complexities of MLMD, focusing on areas such as data management, bias, explainability, and ongoing post-market monitoring. Importantly, this document excludes machine learning medical devices employing large language models (LLM) or generative AI.
Key Topics
-
Risk Management Process
The document outlines how the standard risk management structure set by ISO 14971 should be adapted for ML-enabled medical devices, emphasizing systematic identification, evaluation, and control of risks stemming from AI and machine learning.
-
Roles and Competence
Effective MLMD risk management requires multidisciplinary teams, including expertise in machine learning, clinical practice, human factors, data management, and IT security.
-
Unique MLMD Risks
- Data Quality: Training and test datasets must be representative, robust, and free from bias.
- Bias and Explainability: Properly identifying and mitigating systematic bias is key, as is ensuring AI outputs are interpretable for end-users.
- Autonomy and Oversight: As MLMDs may function with varying levels of independence from human users, controls ensuring appropriate human oversight are emphasized.
- Continuous Learning: Guidance is provided for managing devices that adapt over time, including safe retraining and performance monitoring.
-
Documentation and Traceability
Manufacturers should maintain comprehensive risk management files, capturing the rationale and verification for all risk-related decisions, data processing steps, and software updates.
Applications
-
Development and Testing
The guidance supports the design, development, and validation of ML models in medical devices. Emphasis is placed on establishing acceptance criteria early and ensuring separation of training and testing data.
-
Clinical Integration
By clarifying risk control options and safety measures, manufacturers can facilitate integration of MLMDs into clinical settings while ensuring patient safety and regulatory compliance.
-
Post-Market Surveillance
The standard highlights the importance of monitoring device performance in real-world use, collecting feedback, and updating models as necessary to manage residual risks and maintain device efficacy.
-
Regulatory Compliance
Using this guidance ensures manufacturers meet international expectations for risk management, supporting CE marking and other regulatory approvals in global markets.
Related Standards
-
ISO 14971:2019
Medical devices - Application of risk management to medical devices. The foundational risk management standard for all medical devices.
-
ISO/TR 24971
Provides additional guidance for implementing ISO 14971, especially relevant when dealing with innovative technologies.
-
IEC 80001-1
Application of risk management for IT networks incorporating medical devices.
-
IEC/TR 80002-1
Guidance on the application of ISO 14971 to medical device software.
-
IMDRF N67 and N88
Regulatory documents providing international perspectives on software and AI in medical devices.
Practical Value
CEN ISO/TS 24971-2:2026 is an important resource for organizations developing or evaluating machine learning in medical technology. By building upon internationally recognized standards, it ensures that manufacturers are equipped to address the challenges of AI safety and effectiveness in healthcare. This guidance supports reliable and transparent medical AI deployment, helps reduce patient and user risk, and facilitates compliance with evolving healthcare regulations.
Keywords: machine learning, artificial intelligence, medical devices, risk management, ISO 14971, MLMD, AI bias, data quality, explainability, CEN standards, post-market monitoring, clinical safety.