Secure medical treatment with deep learning on embedded board

Abstract

Deep brain stimulator is among several medical devices known by doctors and scientists for the treatment of movement disorders, such as Parkinson's disease, essential tremor, and dystonia. The security of these devices is the main concern for doctors and patients because any external attacker can introduce fake stimulation inside the human brain and then induce pain or even modify the emotional pattern of the patient. In this chapter, we design a complete prototype of an embedded system for the prediction of different attack patterns in deep brain stimulation (DBS) to mitigate intrusions to such critical devices. We propose the use of the deep-learning methodology to design a deep classifier, based on the dataset obtained from genuine measurements and attack patterns. We prove the robustness of the proposed device by emulating several random attacks on the stimulator. Results show that our system is 97% reliable to predict attacks. We also deploy the proposed system on a cloud and demonstrate the feasibility of detecting the attacks in real time.