FDA has published a 20-page white paper on how to set up certification as a medical device of machine-learning software (SaMD).
FDA document stems from the awareness that technologies based on artificial intelligence (AI) and automatic learning (ML) have the potential to transform healthcare. Examples of high-value applications include early diagnosis, more accurate diagnosis, identification of new models of human physiology, and development of customized diagnoses and therapies. One of the major advantages of AI and ML is its ability to learn from the use and experience of the real world and its ability to improve its performance.
FDA goal is to set up a specific regulatory oversight for this type of software that can improve the effectiveness and quality of care that patients receive.
FDA is developing a regulatory approach to the total life cycle of software as a medical device based on IA/ML. This approach provides that, in the pre-certification of software, FDA can assess the specific company’s culture of quality and organizational excellence, in order to have reasonable assurance of the high quality of software development, testing and monitoring of product performance.
This approach would provide reasonable assurance of safety and efficacy throughout the organisation and its products life cycle, so that patients, healthcare professionals and other users are assured of the safety and quality of these products. This approach allows the evaluation and monitoring of a software from pre-market development to after-sales performance, together with the continuous demonstration of the organization excellence, as shown in the figure below.
To take full advantage of the power of IA/ML learning algorithms, while allowing continuous improvement of their performance and limiting degradation, the FDA approach is based on the following general principles that balance benefits and risks:
- Establish clear expectations on quality systems and good practices for LMs (GMLP);
- Allow manufacturers to submit a change plan during the initial review of an IA/ML-based SaMD before it is placed on the market;
- Expect manufacturers to monitor the IA/ML device and incorporate a risk management approach in the development, validation and execution of algorithm changes (SaMD Pre-Specifications and Algorithm Change Protocol);
- Enable greater transparency to users and FDA through the use of real-world after-sales performance reports to maintain a consistent guarantee of safety and effectiveness.
The document is really very interesting because it addresses and tries to match the need for software certification in a context where software functions and capabilities change during its use. It is a necessary challenge in order not to slow down the research and development of new software.
If you would like to learn more about the subject, please consult the document here.
Digital-health 