Artificial intelligence (A.I.) is an advanced field of science that has influenced various areas. The concept of implementing artificial intelligence in the healthcare has only recently been emerging. The medical field aims to use the power of A.I. to provide more effective and quality treatment to patients by allowing computers to adapt and change continually by gathering data. This will result in better diagnostic, prevention, and treatment of patients.
Basics of Artificial Intelligence and Machine Learning
Artificial intelligence (A.I.) is commonly referred to as an “intelligent” computer programs. Machine learning (M.L.) is a type of artificial intelligence technique which allows computers to improve through sophisticated data algorithms and analysis.
There are two distinct types of machine learning, locked and adaptive.
Locked machine learning is putting limits into the software so that it does not deviate from its original function or the developer’s intention
In contrast, adaptive machine learning allows the computer to change accordingly to the new data and create corresponding new results.
Due to such characteristic, softwares with adaptive machine learning may even develop a new, unintended function.
The FDA has a method of specifically categorizing softwares that are used in the medical field. They are called “SaMD” or “Software as a Medical Device”. SaMD is defined by the IMDRF (The International Medical Device Regulators Forum) as “a software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device”.
Even though artificial intelligence is related to softwares, it is not fully classified as SaMD because of its ability to constantly learn, adapt, and change overtime. The FDA describes softwares using artificial intelligence and machine learning in the medical field as, “AI/ML-based SaMD.”
AI/ML-based SaMD is constantly changing; it “evolves” into a more effective software in correlation with the increased data.
Currently, most of the FDA-approved AI-ML-based SaMD have locked machine learning.
It is difficult to predict how even the locked machine learning will evolve over time by interacting with the real-world data.
The FDA has recently published a proposed regulatory framework regarding AI/ML-based softwares.
Premarket Submission / Modification
A manufacturer may have to submit a premarket submission under the 510(k) software modification guidance if there are changes to the softwares being used and they fall into these categories:
– A change that introduces a new risk or modifies an existing risk that could result in significant harm
– A change to risk controls to prevent significant harm
– A change that significantly affects clinical functionality or performance specifications of the device.
By referencing to the categories set in the 510(k) software modification guidance, it would mean that AI/ML-based SaMD will need premarket submission to the FDA if:
– The modification significantly affects device performance, or safety and effectiveness
– The modification is to the device’s intended use
– The modification introduces a major change to the SaMD algorithm.
PMA (premarket approved) SaMD will also need to submit additional applications for any modifications of the softwares that have new clinical effects, new indication for use that may affect safety or effectiveness, or significant technology modifications that affect performance characteristics.
When is the Premarket Submission Required?
AI/ML-based SaMD is continuously learning through algorithm changes, therefore it may be difficult to set a specific time frame of premarket submission.
FDA states that there must be an assurance of safety and effectiveness while also allowing it to evolve overtime to improve patient care.
AI/ML-based SaMD has vast possibilities of modifications and some of these modifications may not be required to be reviewed according to the guidance provided in “Deciding When to Submit a 510(k) for a Software Change to an Existing Device.”
However, most of the times AI/ML-based SaMD uses new data sets to re-train itself and will require a premarket review.
FDA broadly categorizes these modifications that needs to be reviewed into three categories:
3. The intended use of the SaMD
These modifications can affect one another; the change in performance can influence the intended use, the input can impact the performance, and etc.
Main focus of these modification categories is to ensure that the product does not change from its explicitly claimed performance.
The full discussion paper is available here.
More about FDA