Cyber-Security of Neural Networks in Medical Devices

European and national initiatives, as well as changes in the standards, have led to a dramatic increase in the demand for interconnected medical devices during the last few years. In addition to the increased connectivity of devices in a larger network, economic reasons demand for AI (artificial intelligence) enhanced support functions even in the medical devices domain. Both connectivity and the usage of AI in safety-relevant devices are the basis for new and additional challenges for the manufacturers. The more devices become networked, the more they become targets of cyber-attacks. This means that safety of medical devices can no longer be guaranteed without adequate cyber-security measures. These cyber- security measures must cover the whole design of the devices including their internal neural networks to ensure that the inferences are free from influences from attackers. Modern, advanced attack vectors demand for very advanced countermeasures to ensure or even increase cyber-security of medical devices in this environment. This industrial experience paper will show the required concert of measures that is accompanied by a security informed safety analysis from the earliest steps of development to increase general safety and security in the design of a modern intensive care ventilator.