CRISP-PCCP – A Development Methodology Supporting FDA Approval for Machine Learning Enabled Medical Devices

Abstract

Abstract The U.S. Food and Drug Administration (FDA) is the regulatory body that ensures the safety, efficacy, and security of medical devices and software in the healthcare sector in the U.S. However, its guidelines and regulations often set a global benchmark, influencing medical device standards in Europe and other regions. The FDA recently published a draft guidance, the Predetermined Change Control Plan (PCCP), aiming to support medical device manufacturers with the release of continual learning Machine Learning-Enabled Device Software Functions (ML-DSF). Such ML-DSFs are intended to change after initial market approval. We present a systematic process to support the implementation of the PCCP. Building upon the Cross-Industry Standard Process for the development of Machine Learning applications with Quality assurance methodology (CRISP-ML(Q)), we present an approach that a manufacturer may use to identify and evaluate the impact of anticipated changes to ML-DSF. Our process also indicates a forecast, whether the anticipated change would be accepted by the FDA as a part of the PCCP.

Ähnliche Arbeiten