OnRAMP for Regulating AI in Medical Products

TL;DR

This paper proposes best practice guidelines for developing medical AI solutions, aiming to facilitate regulatory certification and improve communication among stakeholders by aligning machine learning practices with medical device regulation standards.

Contribution

It introduces a risk-based, statistically grounded framework for medical AI development that supports regulatory processes and promotes a common Good Machine Learning Practice (GMLP).

Findings

Guidelines enable clearer communication among developers and regulators.

Framework supports the creation of regulatory packages for medical AI.

Enhances development and regulation of medical AI products.

Abstract

Medical Artificial Intelligence (AI) involves the application of machine learning algorithms to biomedical datasets in order to improve medical practices. Products incorporating medical AI require certification before deployment in most jurisdictions. To date, clear pathways for regulating medical AI are still under development. Below the level of formal pathways lies the actual practice of developing a medical AI solution. This Perspective proposes best practice guidelines for development compatible with the production of a regulatory package which, regardless of the formal regulatory path, will form a core component of a certification process. The approach is predicated on a statistical risk perspective, typical of medical device regulators, and a deep understanding of machine learning methodologies. These guidelines will allow all parties to communicate more clearly in the development of a common Good Machine Learning Practice (GMLP), and thus lead to the enhanced development of both medical AI products and regulations.