FDA’s PCCP: Opportunities and Gaps

Niklas Babendererde, Amin Ranem, Moritz Fuchs, Camila Gonzalez, Henry John Krumb, Anirban Mukhopadhyay

Keywords: PCCP, Continual Learning, Medical Imaging

TL;DR: Potential and issues of PCCP, along with suggested solutions to these issues

Abstract:

The dynamic nature of medical imaging data poses a significant regulatory challenge for AI-based Software as a Medical Device (SaMD), as it requires constant adaptation. Traditionally, each of these modifications would require the SaMD to go through the complete approval process again, limiting the real-world deployment of AI-assisted SaMDs. The U.S. FDA's Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (PCCP) aims to bridge this gap. It allows, under certain conditions, a significantly simplified approval process for updated AI-enabled SaMDs. In this work, we discuss the great potential that this brings for the dynamic reality of medical imaging, but also explain three potential gaps in the current regulation. These concern an "evaluation gap" that poses a potential loophole from modified test data, an "intended use gap" that limits flexibility for unexpected and time-critical events such as a pandemic, and the "foundation model gap" that limits the applicability of this emerging technology.
For each of those gaps, we present a solution to fully leverage the potential of PCCP as a regulatory framework enabling technologies that address the dynamic reality of medical imaging.

Submission Number: 4