Parametric heart modelling and in silico closed-loop validation of implantable devices

Abstract

Medical Cyber-Physical systems (MCPS), such as cardiac pacemakers and implantable cardioverter defibrillators (ICDs), are life-critical and context-aware. These devices have been increasingly used to provide continuous health-care for patients. The complicated heterogeneous physical dynamics of patients brings distinct challenges for the device development and validation. Testing implantable devices in a closed-physiologic-loop with the actual organ can provide key insights into design safety and treatment efficacy, but this is rarely practical or comprehensive. An in silico closed-loop validation platform allowing interactive and clinically relevant evaluation is desirable. The in silico closed-loop system, incorporating a physiological environment model, is likely to reveal safety risks at the early stage of development before animal testing and clinical trials. However, the closed-loop validation of implantable cardiac device is yet to be thoroughly investigated. Hence, this research proposes three key methodologies to address this need: heart modelling, electrogram (EGM) modelling and automation of closed-loop validation. In an ideal closed-loop system, a heart model responds to the device outputs in realtime according to the underlying physiological dynamics. Although biophysically detailed models provide accurate and realistic insights with predictive dynamics, they are unable to respond in real-time due to the expensive computational cost. While existing simplified models can provide real-time responses, they may not have the necessary dynamic features. We address this problem with a new high-level (abstracted) physiologically based computational heart model. The model comprises a network of nodes along the cardiac conduction system. Each node represents distinct regional electrical action potential behaviour. These nodes are connected by path models which encode the biophysics of tissue conduction. The regional electrophysiology and paths are modelled with hybrid automata (HA). The HA-based model can capture continuous non-linear electrophysiological dynamics while retaining computational efficiency with the possibility of formal analysis. The hierarchy of pacemaker functions is incorporated to generate sinus rhythms, while abnormal automaticity can induce a variety of arrhythmias such as escape ectopic rhythms. Model parameters are calibrated using experimental data and prior simulations. Regional electrophysiology and paths in the model can match human action potentials, dynamic behaviour, and cardiac activation sequences. The implanted cardiac device will sense a complex combination of near and far-field extracellular potentials together with noncardiac signals-electrograms (EGMs). All modern implantable devices incorporate blanking periods and refractory periods to discriminate these signals and identify the intrinsic local depolarization. To the best of our knowledge, no existing work explicitly models far-field signals and pacing artifacts for a virtual heart. We develop an intracardiac EGM model as an interface between the virtual heart and the physical device. The model, based on classical dipole theory, can capture not only the local excitation but also far-field signals and pacing afterpotentials. This significantly extends the capabilities of the virtual heart model in the context of device validation. The same HA formalism is employed such that the EGM model can be seamlessly integrated into the HA-based virtual heart. The functioning of the virtual heart relies on the appropriate selection of values for hundreds of parameters. The relationship between the model parameters and the target physiological space is not straightforward. An exhaustive exploration of the high dimensional system is infeasible. This becomes more challenging in the closed-loop context, where the heart behaviour is tightly coupled with the device. We propose a closed-loop framework to facilitate the automation of the model parametrisation. This framework includes three components: test generation, execution, and evaluation. These form another closedloop system. The engine of the closed-loop system is a stochastic optimization tool, which guides the parameter search space towards a region of interest. As a consequence, the heart model exhibits various arrhythmias and provides an intended input spectrum to the device. We illustrate the application of the cardiac physiological model in requirements-based testing, device safety risk evaluation, and device programming customisation. The simulated findings agree well with clinical observations.