Medical Device Safety: An Analytical International Survey of AI/ML-Enabled Medical Devices

This paper analyzes gaps in key safety issues for U.S. Food and Drug Administration and international approvals of artificial intelligence/machine learning-enabled medical devices (AIMD). National Institute of Standards and Technology (NIST) provides guidance for patient safety and AIMD security. For example, NIST Artificial Intelligence Safety Institute, NIST Special Publication (SP) 800-82r3, and NIST SP 1500-202, Section 2.3.3, suggest approximate patient safety analysis approaches. Related NIST security issues are identified in NIST guidance for artificial intelligence/machine learning risk management, such as NIST's adversarial machine learning guidance. An analytical literature review was conducted within the Johns Hopkins University Sheridan Libraries Catalyst databases to support stated objectives around selected NIST common language terms related to safety and security. The resulting analysis demonstrates that, in the U.S. and on an international scale, selected AIMD literature and guidance may not include or consistently use critical, integrated terms for AIMD security or safety, such as NISTIR 8312's four principles of artificial intelligence explainability. In conclusion, AIMD safety and security could be addressed with NIST common languages.