Another in a series of previews for Sensors Converge 2023, June 20-22 in Santa Clara, California
When developing a medical device that utilizes machine learning, there are critical elements to understand and incorporate. Here are a few that companies should be considering, especially when pursuing U.S. Food and Drug Administration certification. (Adapted from a collaboration with Dojo Five and Edge Impulse; read the full article here.)
Validation and documentation
The importance of validation and documentation in embedded firmware when integrating ML into a medical device cannot be stressed enough. The documentation of device specifications is critical for the regulatory submission process.
The FDA requires detailed documentation of the device's intended use, performance characteristics, and operating conditions as part of the premarket approval process. Documentation is vital in the regulatory submission process and the FDA has published guiding principles for medical devices with machine learning.
Device classifications and testing requirements
There are three FDA classifications for medical devices; companies need to know which classification a device falls under.
Class I: A medical device with low to moderate risk that requires general controls such as establishment registration, quality system regulation, and labeling requirements
Class II: A medical device with a moderate to high risk that requires special controls. These devices require general controls and special controls, which may include performance standards, guidelines, and testing requirements. Most Class II devices need premarket notification (510(k) clearance) to demonstrate their safety and effectiveness.
Class III: A medical device with high risk that requires premarket approval. These devices must comply with general controls, special controls, and premarket approval (PMA) requirements. PMAs involve extensive testing, including preclinical and clinical trials, to demonstrate the device's safety and effectiveness.
Various types of testing are required for FDA approval:
- Biocompatibility testing assesses the device's biological response to ensure it is safe for use in, or on, the human body.
- Electrical safety and electromagnetic compatibility testing evaluate the device's safe operation and interaction with other electrical equipment.
- Performance testing examines the device's accuracy and functionality, while software validation and verification testing ensure the software's safety and effectiveness
- Usability testing focuses on the device's ease of use and user experience.
[More on testing requirements]
Training, data, and documentation
When developing a medical device with ML, it is crucial to define clear specifications, including the algorithms used, data sources, and limitations. The training data should be representative of the intended patient population, and potential biases must be addressed. Rigorous validation, potentially using independent datasets, is necessary to demonstrate the device's generalizability.
Documentation requirements for FDA approval depend on the device classification and chosen regulatory pathway. The FDA requires detailed documentation of the device's intended use, performance characteristics, operating conditions, and ML algorithms used. This documentation must outline the data sources, training, and validation processes. [Read our blog on Medical Device Documentation]
Guardrails
Embedded software is a critical component of many modern medical devices and is responsible for controlling the device's operations, processing data, and generating outputs. Guardrails must be implemented to ensure that the outputs generated by the ML algorithm are within the expected range of values and comply with regulatory requirements. Testing the guardrails is essential to ensure patient safety and the proper functioning of the medical device.
In conclusion, the integration of machine learning (ML) into medical devices requires special attention to testing, regulatory compliance, and the specific considerations related to ML development. Overall, following these considerations and requirements can help in the successful development and approval of medical devices with integrated machine learning.
Cutline for the image: Integration of machine learning (ML) into medical devices adobe license
Tom Dever is vice president of product at Dojo Five and Mike Senese is director of content marketing at Edge Impulse. Dojo Five helps medical device customers improve the reliability, quality, and consistency of their embedded development projects. Learn more about their modern tools, techniques, and best practices for embedded medical device development projects.
Special note: Dever is speaking at Sensors Converge on Thursday at 1:30 p.m. PT on “Integrating TinyML into your Modern Embedded Firmware Development Workflow.” Dojo Five can be found at booth 702 during the event, running June 20-22 in Santa Clara, California. Registration is online.