Characterizing sensor accuracy requirements in an artificial intelligence-enabled medical device

•Inaccurate sensor inputs affect score outputs in AI-enabled devices.•Simulation-based method can predict how sensor inaccuracy affects algorithmic output.•Simulation-based method can be used to determine design requirements.•Ensuring algorithmic outputs are accurate is important for reducing bias in AI. Artificial intelligence and machine learning applications are increasingly prevalent in the healthcare industry. In some cases, medical devices use sensor-collected data to feed into algorithms which generate scores or risk assessments that are used to inform patient care. The process of determining sensor accuracy requirements which will ensure that the algorithm generates reliable scores is not straightforward or well-defined. In this paper, we describe a simulation-based method to characterize sensor accuracy requirements for a device that uses a machine-learning algorithm to generate a postural stability score – the ZIBRIO Stability Scale. The results of the simulation are described, as is the application to sensor selection in preparation for manufacturing of the device. Other medical device developers may be able to use this method or similar methods in their requirements engineering process.