| Autor: |
González, Camila, Fuchs, Moritz, Santos, Daniel Pinto dos, Matthies, Philipp, Trenz, Manuel, Grüning, Maximilian, Chaudhari, Akshay, Larson, David B., Othman, Ahmed, Kim, Moon, Nensa, Felix, Mukhopadhyay, Anirban |
| Rok vydání: |
2024 |
| Předmět: |
|
| Druh dokumentu: |
Working Paper |
| Popis: |
Over time, the distribution of medical image data drifts due to multiple factors, including shifts in patient demographics, acquisition devices, and disease manifestation. While human radiologists can extrapolate their knowledge to such changes, AI systems cannot. In fact, deep learning models are highly susceptible to even slight variations in image characteristics. Therefore, manufacturers must update their models with new data to ensure that they remain safe and effective. Until recently, conducting such model updates in the USA and European Union meant applying for re-approval. Given the time and monetary costs associated with these processes, updates were infrequent, and obsolete systems continued functioning for too long. During 2024, several developments in the regulatory frameworks of these regions have taken place that promise to streamline the process of rolling out model updates safely: The European Artificial Intelligence Act came into effect last August, and the Food and Drug Administration (FDA) released the final marketing submission recommendations for a Predetermined Change Control Plan (PCCP) in December. We give an overview of the requirements and objectives of recent regulatory efforts and summarize the building blocks needed for successfully deploying dynamic systems. At the center of these pieces of regulation - and as prerequisites for manufacturers to conduct model updates without re-approval - are the need to describe the data collection and re-training processes and to establish real-world quality monitoring mechanisms. |
| Databáze: |
arXiv |
| Externí odkaz: |
|
