Securing Insulin Pump System Using Deep Learning and Gesture Recognition

Modern medical devices are equipped with radio communication chips enabling medical practitioners to remotely and continuously monitor patient's health. The conjunction of these medical devices with the radio communication chips and their internet connectivity exposes them to security and privacy risks. The insulin pump system is an autonomous, wearable external device, commonly used by diabetic patients to take insulin efficiently, as compared to manual injection through a syringe. Security attacks may disrupt the working of insulin pump system by delivering the lethal dose to patients and endanger their lives. In this paper, we ensure the correct dosing process of insulin pump system based on the combination of deep learning model and gestures performed by the patient. Specifically, we used Long Short-Term Memory (LSTM) recurrent neural network to predict the thresh hold value of insulin based on last three months log of insulin pump system. If the amount of insulin to be injected by the insulin pump system is greater than our predicted thresh hold amount, then our system asks the patient to perform the gesture. After successful recognition of the patient's gesture, our solution compares the suspicious value of insulin with patient's gesture and identifies an attack.