Scribd
Upload
11 views

Good Machine Learning Practice

IMDRF Machine Learning

Uploaded by

martyna.piecko
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
11 views

Good Machine Learning Practice

IMDRF Machine Learning

Uploaded by

martyna.piecko
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

14:45 – 15:00

Good Machine Learning


Practice (USA / UK)

Melissa Torres
Disclaimer
Associate Director, U.S. Food and Drug Administration
This document was produced by the International Medical Dev ice Regulators Forum. There are no restrictions on the reproduction or use of this document;
howev er, incorporation of this document, in part or in whole, into another document, or its translation into languages other than English, does not conv ey or
represent an endorsement of any kind by the International Medical Dev ice Regulators Forum.
Copy right 2021 by the International Medical Dev ice Regulators Forum.
ARTIFICIAL INTELLIGENCE (AI)/MACHINE
LEARNING (ML) ENABLED MEDICAL DEVICES
WORKING GROUP UPDATE
Co-Chairs:
Matthew Diamond – US FDA
Russell Pearson – UK MHRA
March 2023

Background

Good Machine Learning Practice (GMLP) brings together high-level,


fundamental principles important for the development, use and
monitoring of Machine Learning (ML)-enabled medical devices.

ML-enabled products have unique considerations that can be


addressed, at least in part, with GMLP implemented across the
product life cycle.

Rapid technological advancements in AI/ML, combined with


manufacturers from sectors beyond medical devices, makes
development of GMLP an important priority to lower product and
development risks and to protect against regulatory divergence.

220
March 2023

Rationale

• There is a close interplay between ML-enabled medical devices and other software
based medical devices.
– GMLP must be developed to be compatible with best practice SaMD guidance and built
upon existing core aspects of medical device regulations.
○ Core aspects include quality management systems, risk management and clinical
evaluation to ensure the GMLP complements the state of the art and regulatory
compliance.
• Work is aligned with the IMDRF Strategic Plan to develop a harmonized approach to
the management of AI medical devices.
• Generating consensus across the product lifecycle via creation of a GMLP document
will also assist with the IMDRF objectives to strengthen post-market activities and a
total product lifecycle regulatory approach to medical devices.

221
March 2023

Goal

• To develop a new document on the topic of Good Machine Learning Practice


(GMLP) that will provide internationally harmonized principles to help promote
the development of safe and effective ML-enabled medical devices.
– Generate consensus on the key considerations for how AI/ML products can meet
regulatory, risk management, quality management and clinical evaluation in order to
promote consistency across jurisdictions.
– Build upon the US FDA/Health Canada/UK MHRA joint document on GMLP Guiding
Principles.

222
March 2023

Good Machine Learning Practice Principles

Good Machine Learning Practice for Medical Device Development:


These guiding principles may be used to: Guiding Principles
• Adopt good practices that have been Multi-Disciplinary Expertise are Leveraged Good Software Engineering and Security Practices are
proven in other sectors; Throughout the Total Product Life Cycle Implemented

Clinical Study Participants and Data Sets are


• Tailor practices from other sectors so they Training Data Sets are Independent of Test Sets
Representative of the Intended Population
are applicable to medical technology and
Selected Reference Datasets are Based Upon Best Model Design is Tailored to the Available Data and
the health care sector; and Available Methods Reflects the Intended Use of the Device
• Create new practices specific for medical Focus is Placed on the Performance of the Testing Demonstrates Device Performance during
technology and the health care sector. Human-AI Team Clinically Relevant Conditions

Deployed Models are Monitored for Performance and


Users are Provided Clear, Essential Information
Re-training Risks are Managed

223
March 2023

Current Status

• New Work Item Proposal approved in January 2022


• Working group is currently being established
– Call for participants/representatives from IMDRF regulatory authorities, RHIs,
and industry
– IMDRF website updates
– Expect work to begin in the next couple of weeks

224
Thank you!
Questions?
Email [email protected]
[email protected]

Disclaimer
This document was produced by the International Medical Dev ice Regulators Forum. There are no restrictions on the reproduction or use of this document;
howev er, incorporation of this document, in part or in whole, into another document, or its translation into languages otherthan English, does not conv ey or
represent an endorsement of any kind by the International Medical Dev ice Regulators Forum.
Copy right 2021 by the International Medical Dev ice Regulators Forum.

You might also like