The Artificial Intelligence and Machine Learning (AI/ML) Program in the FDA’s Center for Devices and Radiological Health (CDRH) conducts regulatory science research to ensure patient access to safe and effective medical devices using AI/ML. This is one of 20 research programs in CDRH’s Office of Science and Engineering Laboratories (OSEL).
Artificial Intelligence, Machine Learning, and Medical Devices
The growth of commercial AI/ML-based technologies has shown a spillover of AI/ML for use as and within medical devices, with important contributions in application areas such as:
- Image acquisition and processing
- Earlier disease detection
- More accurate diagnosis, prognosis, and risk assessment
- New patterns identification on human physiology
- Personalized diagnostics and therapeutics.
Challenges in developing robust clinical and non-clinical evaluation methods and in better understanding the effects of AI/ML devices in the real world stem from factors such as:
- The rapid application of AI/ML technology
- The unique nature of clinical medical data (for example, low prevalence of disease, lack of or difficulty in obtaining truth data, and so forth).
In addition, the ability of AI/ML for continuous learning presents unique regulatory challenges with a need to develop appropriate regulatory controls and test methods to balance the potential benefits of AI/ML adaption with the risks of a limited assessment paradigm.
Regulatory Science Gaps and Challenges
Major regulatory science gaps and challenges that drive the Artificial Intelligence and Machine Learning Program are:
- Lack of methods that can enhance AI/ML algorithm training for clinical datasets that are typically much smaller than non-clinical datasets.
- Lack of clear definition or understanding of artifacts, limitations, and failure modes for fast-growing applications of Deep-Learning (DL) algorithms in the denoising and reconstruction of medical images.
- Lack of a clear reference standard for assessing accuracy of AI/ML-based Quantitative Imaging (QI) and radiomics tools.
- Lack of assessment techniques to evaluate the trustworthiness of adaptive and autonomous AI/ML devices (for example, continuously learning algorithms).
- Lack of systematic approaches to address the robustness of various AI/ML input factors, such as data acquisition factors, patient demographics, and disease factors, to patient outcomes in a regulatory submission.
The Artificial Intelligence and Machine Learning Program is intended to fill these knowledge gaps by developing robust AI/ML test methods and evaluating test methodologies for assessing AI/ML performance both in premarket and real-world settings to reasonably ensure the safety and effectiveness of novel AI/ML algorithms.
Artificial Intelligence and Machine Learning Program Activities
The Artificial Intelligence and Machine Learning Program focuses on regulatory science research in these areas:
- Data augmentation, transferring learning, and other novel approaches to enhance AI/ML training/testing for small clinical datasets.
- Study design and analysis methods for AI/ML-based computer-aided triage (CADt).
- Non-clinical phantoms and test methods for assessing specific imaging performance claims for DL-based denoising and image reconstruction algorithms.
- Imaging phantoms and computational models to support QI and radiomics assessment.
- Assessment techniques for evaluating the reliability of adaptive AI/ML algorithms to support non-clinical test method development.
- Assessment approaches to estimate and report the robustness of AI/ML to variation in data acquisition factors.
For more information, email OSEL_AIML@fda.hhs.gov