Medical Cyber-Physical Systems (CPS) diagnose the physiological conditions of the patients and autonomously deliver therapy with little or no human intervention. The autonomy enables timely therapy and better lifestyle for the patients, but also presents unique challenges to the safety and efficacy of medical CPS. Unlike CPS in other domains where the systems only interact with the human-made physical environment, medical CPS interact with human physiology which is less understood with much larger variability. It is impossible to consider all possible physiological conditions during system design, therefore there may exist rare conditions in which the system operates as designed but not as intended, causing serious injury or even death of the patient. In this research, we used implantable cardiac devices as example to demonstrate the use of physiological modeling in closed-loop validation of implantable cardiac devices (Proceedings of IEEE'12). The physiological models enabled formal techniques like model checking to be applied to provide safety assurance of the devices (STTT'14), and can be seamlessly integrated into the development and certification process (IEEE Computer'16).
The safety and efficacy of the closed-loop devices should be evaluated within their physiological context. In this work we developed a heart model structure which can be used for closed-loop validation of implantable cardiac devices. The model structure is abstract enough for model checking, while expressive enough to model various heart conditions.
Model checking can be used to verify whether medical device software model can safely operates within all possible physiological contexts. The key challenge is to 1) ensure validation efficiancy, and 2) provide interpretability for counter-examples. In this project we study the abstractions and refinements of physiological models during closed-loop model checking, in order to balance between model coverage and model expressiveness.
Currently, Cardiac Ablation is the primary method to treat cardiac arrhythmia. The condition of the heart can be diagnosed by analyzing the timing and patterns of electrical events sensed by the electrodes inserted into the heart. The accuracy and time consumption of the diagnosis depends on the experience of the physician, which varies within a huge range. In this project, we propose a clinical support system to improve the accuracy and efficiency of cardiac ablation procedures. The system takes real-time electrical signals and location of the electrodes as input, and provide the physicians with 1) annotated electrical signals; 2) possible heart conditions given input history; and 3) guidance towards the ablation sites.
Clinical trials are the ultimate closed-loop validation of the medical devices, which are expensive and prone to failures. In this project, we propose in silico pre-clinical trials, in which the devices are evaluated on a virtual patient population consists of physiological models. The results of these in silico trials can be used to provide insights that can be beneficial to the actual clinical trials.
Driving is a social activity which involves endless interactions with other agents on the road. Failing to know where these agents are and predict what they will do may cause serious safety hazards. With more and better sensors on board, modern vehicles can collect historical behaviors of the vehicles. These behaviors can be used to construct behavioral models of the drivers, which can be used to predict their future behaviors under different driving contexts. With connected technology, these models can be shared among connected vehicles to better predict behaviors of each other, which can be used to predict potential collisions.
2018 © Zhihao Jiang. ALL Rights Reserved.