Importance to FDA

Artificial intelligence (AI) solutions have the potential to improve automation and learning of medical devices, the efficiency of diagnostic/therapeutic development and commercial manufacturing, regulatory assessment, and postmarket surveillance, among many other potential applications. These improvements increase the accuracy of predictive modeling, enable efficient automation of medical devices and manufacturing processes, leverage knowledge management resources to improve regulatory review, and focus and improve postmarket surveillance. FDA views AI as encompassing continued improvement in code and infrastructure.

To achieve and promote efficiencies within FDA and in industry, FDA aims to improve its understanding of AI’s potential and limitations. Considerations include the technical and practical application of AI, new regulatory questions introduced by AI applications, and the impact of AI solutions across the lifecycle of FDA-regulated products.

Examples

FDA advances the understanding and use of AI to support a diverse set of needs related to FDA-regulated products, including:

• Exploring the use of machine learning (ML) algorithms to:
  • Target high-risk seafood products offered for import;
  • Detect adverse events in different data sets, including post market data; and
  • Study the effects of using synthesized data sets for training and testing in both pre-market testing and the FDA-regulated product lifecycle; and
  • Predict the time to first submissions for abbreviated new drug applications (ANDA) referencing new chemical entities to inform the Agency’s ANDA workload and prioritize research.
• Evaluating the potential impact of natural language processing (NLP) systems to automatically identify and International Council for Harmonization Medical Dictionary for Regulatory Activities (MedDRA) code adverse events mentioned in product labels. The labeled status of the MedDRA-coded adverse events in product labels facilitate the triaging and review of the adverse events described in individual case safety reports (ICSRs) submitted to the FDA Adverse Event Reporting System (FAERS).
• Conducting research on AI/ML-based medical devices:
  • 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.
  • Technical factors influencing AI reproducibility for digital pathology applications.
  • Methods for assessing the generalizability of AI performance in digital pathology applications.
• Investigating the potential of AI to improve the efficiency of reviewing regulatory submissions. For example, FDA applies natural language processing to regulatory submissions to classify its relative complexity.
• Studying how AI can combine diverse data so clinical trial results can be analyzed in a more comprehensive and expeditious way.
• Applying AI to support the processing and evaluation of individual case safety reports submitted to FAERS and VAERS by leveraging NLP, ML, and visualization techniques to facilitate efficient FAERS data analyses.
• Exploring how AI can be used in pharmacometrics, the science that quantifies drug, disease, and trial information, to aid efficient drug development, and/or regulatory decisions.
• Exploring how AI can be used to advance precision medicine, by predicting patient responses based on baseline patient characteristics.
• Developing standardized ontology-based metadata to leverage WGS data to predict source tracking regarding domestic animal host of food source. These metadata ontologies also support risk assessment tools such as GenomeGraphR, a user-friendly open-source web application for foodborne pathogen WGS data integration, analysis, and visualization.
• Designing and delivering workshops to introduce AI and ML to FDA staff and showcase how AI and ML can be applied to FDA’s regulatory landscape as the implementation of these technologies becomes ubiquitous across the spectrum of regulated industry. Workshops such as this enhance the knowledge base of our inspectional cadre, keeping them attuned to how these emerging technologies may impact their investigations.
• Using ML to develop computer models that use genomic data to predict the mean inhibitory concentration (MIC) for pathogens and antimicrobials surveyed for the National Antimicrobial Resistance Monitoring System (NARMS). The goal is to develop reliable methods to predict MICs from whole-genome sequence data.
• Conducting educational Al and ML workshop through the CERSI program on potential causal inference, focusing on heterogeneous effects, complex data structures, interpretability and explainability of AI results.
• Researching image blending to expedite development and performance assessment of mammographic computer-aided diagnostic (CAD) devices. The goal of this work was to reduce the data burden for medical device manufacturers and to provide information to inform review of mammographic CAD systems.
• Designing a statistical framework that would be robust to distributional shifts over time for software as a medical device (SaMD) adapting in the real world to ensure the safety and effectiveness of potential artificial intelligence/machine learning (AI/ML)-based SaMDs under a CERSI research project.
• Developing a framework with a CERSI partner for measuring robustness of ML models to contextual changes in the real-world data and inform regulatory decision-making by categorizing which contextual factors matter for a particular intended use and how to better define the context of appropriate use.
• Exploring how AI is used during commercial drug manufacturing to improve quality decision making.