Validation of AI/ML diagnostic in view of regulatory compliance

Interested, who to contact?

@Vladimir Makarov

The innovation team: projectenquiries@pistoiaalliance.org

The Problem

Unlike traditional diagnostics and medical devices, AI and ML systems have an ability to continuously learn based on exposure to new patients and data in clinical settings and hence their performance may vary over time. This represents a unique challenge to regulators. Multiple other issues may also be challenging: availability and use of data, public trust, algorithmic bias, accountability, transparency, reproducibility and explainability of AI algorithms. Recently both the European Commission and FDA have released draft AI ethics and regulatory guidelines in early 2019 (FDA Proposal April 2019).

The challenge as an industry is to collaborate as a group, understand the implications and align with the regulators.

The Proposal:

Setup a community of interest with teleconferences (and if possible in-person meetings).
Work with regulators in NA and Europe to begin the discussions.
Publish opinion pieces and white papers and develop best practices.

This proposal has a sister one at https://pistoiaalliance.atlassian.net/wiki/spaces/IC/pages/1876197464 A community of interest has recently formed there.