Artificial Intelligence (AI) and Machine Learning(ML) have experienced substantial growth across multiple industries, including healthcare. The integration of these progressive technologies into medical devices already shows promise in detecting infection-related complications swiftly and accurately. However, this integration brings along challenges in aligning it with the established regulatory norms, adding the complexity of devices that provide unpredictable outputs. Consequently, specific medical devices employing AI/ML technology demand a tailor-fit regulatory process for their market acceptance.
By March 25, 2025, AI-based software had achieved considerable ground in the medical device industry. The U.S Food and Drug Administration (FDA) approved more than 1015 devices with AI/ML software, including the innovative Sepsis ImmunoScore. This FDA-authorized AI software, approved in April 2024, was designed to assist in predicting and diagnosing sepsis. It uses AI/ML software to extract data from the electronic medical record that specifies the patient’s risk of developing sepsis within 24 hours. Reliability notwithstanding, regulatory bodies must address the trustworthiness of these AI/ML devices.
Understanding the differing AI software types is paramount in acknowledging the need for tightened regulation over devices using these systems. Data-driven ML systems like locked ML models require external approval for changes, differing from continuous learning models, which self-update their algorithms. The latter becomes an issue from the regulatory standpoint due to the unpredictability of algorithm shifts. The introduction of AI in the medical device sector, especially continuous learning models, has aroused concerns over trust with tasks traditionally performed by humans. Furthermore, it has raised questions about its predictability and reliability.
The FDA identifies three main types of software in the medical device industry: Software as a Medical Device (SaMD), software used in medical devices, and software used in manufacturing. SaMD includes AI/ML software aimed at diagnosing, mitigating, curing, or preventing disease or other conditions. SaMD, along with other forms of medical devices, is categorized by FDA into three different classes according to the associated risk to the patient. Class I represents low risk, while Class III indicates the greatest risk. Depending on the category, different premarket submission requirements are necessary. The class III devices, given their high-risk nature, require a premarket approval application (PMA) that provides evidence of the device’s safety.
Given the complex nature of AI/ML technology, the application of this will range widely in the domain of future medical devices. It is necessary for healthcare professionals to continue advocating for a robust regulatory framework for these devices. Embracing evolving technology while ensuring its safe and reliable use demands this advocacy for tailoring the regulatory process to these devices.