FDA Classifies Radiological Machine Learning-Based Quantitative Imaging Software with Predetermined Change Control Plans as Class II Devices
The U.S. Food and Drug Administration (FDA) has published a final order classifying Radiological Machine Learning-Based Quantitative Imaging Software with Predetermined Change Control Plans (PCCPs) as Class II medical devices under 21 CFR §892.2055.
The new classification establishes a dedicated regulatory framework for software-only medical devices that use machine learning algorithms to analyse radiological images and generate quantitative imaging outputs. The final order also introduces special controls intended to provide reasonable assurance of the safety and effectiveness of these devices while supporting innovation through the De Novo classification pathway.
New Classification for AI-Based Imaging Software
The FDA explains that the new device type covers software that applies machine learning algorithms to radiological images to generate quantitative imaging outputs.
Examples of supported functions include:
View selection;
Image segmentation;
Landmark detection.
The regulation also applies to devices designed with a Predetermined Change Control Plan (PCCP), allowing specific future software modifications to be implemented under an established regulatory framework.
Special Controls Introduced by FDA
To support the Class II classification, the FDA has established a series of special controls applicable to this new device type.
These include requirements for:
Design verification and validation;
Documentation of training datasets and algorithm development;
Performance testing using independent validation datasets;
Software verification, validation and hazard analysis;
Risk management for planned software modifications;
Comprehensive device labelling.
The agency also requires manufacturers to document planned software modifications covered by the PCCP together with the associated verification and validation methodology.
Requirements for Device Labelling
The final rule specifies detailed labelling requirements for manufacturers.
According to the FDA, labelling should include information such as:
The patient population used during validation;
Intended users and required expertise;
Device inputs and outputs;
Compatible imaging hardware and imaging protocols;
Performance testing results;
Known performance limitations;
Information relating to the Predetermined Change Control Plan, including implemented modifications and version history.
Supporting Innovation Through the De Novo Pathway
The FDA notes that the device was originally reviewed through the De Novo classification process, allowing the agency to establish a new Class II device type with special controls.
According to the agency, this approach supports innovation by enabling future substantially equivalent devices to use the 510(k) pathway rather than requiring a new De Novo request, potentially reducing regulatory burden while maintaining appropriate regulatory oversight.
Relevance for Medical Device Manufacturers
The final order is particularly relevant for manufacturers developing artificial intelligence and machine learning-based medical imaging software intended for the U.S. market.
Key topics addressed include:
AI-enabled radiological software;
Predetermined Change Control Plans (PCCPs);
De Novo device classification;
Design verification and validation;
Software lifecycle management;
Machine learning performance evaluation;
Regulatory requirements for Class II medical devices.
The publication provides further insight into FDA's evolving regulatory approach for AI-enabled medical devices and the use of PCCPs to support controlled software modifications throughout the product lifecycle.
