A Deep Learning Approach for ECG Authentication on Implantable Medical Devices

Abstract

The rise in smart healthcare systems have enhanced connectivity of implantable medical devices (IMD). Healthcare providers are able to wirelessly control and monitor these devices enabling quicker diagnosis and treatment. However, the devices' underlying wireless communication medium also pose security risks for patients, as unauthorized access could result in exposing private information and compromising the devices critical functionality. In this work, we develop a biometric based authentication using deep learning for entities seeking access to IMDs. Specifically, we utilize the patients Electrocardiogram (ECG) signal to authenticate programmers attempting to communicate with the IMD. We implement varying neural network models and evaluate them based on their authentication accuracy. Simulation results show that CNN model with 10 hidden layers performed best with 99.7% accuracy.