Targeted adjustments are also made to the relevant QMS processes:
- Strategy for regulatory compliance,
- Personnel,
- Manufacturer information,
- Infrastructure,
- Purchasing and suppliers,
- External processes, and
- Information security/data protection.
Extension of software lifecycle processes to the design and development of AI models
Manufacturers of AI-based medical devices apply the EN 62304 standard to the software life cycle. If the software is a standalone product, the EN 82304-1 standard is also relevant. To this end, the manufacturer implements a development process, a release process, a maintenance process, and a decommissioning process for the software within its QMS. These processes are supplemented by the manufacturer’s validation and post-market activities.
BAIM supplements software development in accordance with EN 62304 with a subordinate AI-specific development process. While the collection and analysis of all users, system, and AI requirements take place within the regular software development process, data management and AI model development are carried out in a separate AI development process. The review and validation of the AI model then take place in an additional AI model evaluation process.
Adaptation of other key QMS processes
A risk management system is mandatory as part of the quality management system for manufacturers of medical devices. Since AI-based technologies entail specific risks such as bias or misinterpretation of data, these must be specifically addressed in risk management. BAIM therefore expands the traditional risk analysis to include AI-specific hazards, thereby supporting the safe design and development of AI-based medical devices.
In addition, BAIM takes a holistic approach to risk management in accordance with EN ISO 14971 by jointly reviewing safety, cybersecurity, and their interactions. To this end, BAIM utilizes the VDE’s ARGOS cybersecurity approach to systematically identify AI-specific assets, interfaces, and potential attack scenarios and to derive appropriate protective measures.
In the usability engineering process, both formative reviews during design and development and maintenance and summative validations at the conclusion of the development process are conducted. BAIM supports this process through an integrated checklist with specific requirements for the usability of AI-based medical devices. Among other things, this considers topics such as transparency, the explainability of AI decisions, and the handling of automation bias.
The clinical evaluation serves to demonstrate the safety, performance, and a positive benefit-risk ratio of a medical device. In addition to the established requirements of MEDDEV 2.7/1 and Guideline MDCG 2020-1 for medical software, BAIM supports this process with an AI-specific checklist. This checklist addresses additional considerations in the review of AI-based medical devices and takes into account, for example, the use of specialized AI databases for scientific literature searches.
Post-market surveillance and vigilance are also essential components of the software life cycle. Continuous performance monitoring and ensuring product quality in real-world use are of particular importance, especially for AI-based medical devices. BAIM therefore expands the post-market surveillance process to include AI-specific aspects, such as the review of the quality of real-world data and the up-to-date status of the ground truth or gold standard used. To this end, suitable methods, inspection intervals, and relevant metrics and thresholds are defined and integrated into the surveillance process.
Practical implementation
A typical BAIM project begins with defining the intended purpose of the AI-based medical device, considering the AI technology employed and its operating principle. Building on this, the AI-specific development processes are integrated into the QMS, and the necessary risk management activities are established. At the same time, usability measures are planned and carried out, and the clinical evaluation is initiated. After the product is placed on the market, the performance of the AI-based software is continuously monitored as part of post-market surveillance, clinical follow-up, and vigilance. The insights gained from this process are continuously incorporated into the clinical evaluation and further development of the product.
Conclusion
CE conformity for AI-based medical software requires systematic consideration of regulatory, technical, and clinical requirements. The BAIM approach supports manufacturers in the structured implementation of these requirements and in ensuring that the evidence chain is traceable throughout the entire product lifecycle.
BAIM can both expand an existing QMS and assist in its implementation. Using BAIM ensures that the Manufacturer’s QMS is prepared for new regulatory challenges.
Are you developing AI-based medical devices and looking to efficiently meet regulatory requirements? Feel free to contact us to learn how BAIM can support you on your path to CE Conformity.
