1st Supervisor: Dr Martin Bishop, King’s College London
2nd Supervisor: Dr Andrew King, King’s College London
Clinical Supervisor: Dr John Whittaker, Guy’s & St Thomas’ NHS Foundation Trust, King’s College London
Aims of the Project:
- To use an in-silico approach, enabled by AI technology, to personalise the implantation, sensing, decision-making and electrotherapy protocols of implanted cardioverter defibrillators to enhance efficacy and safety.
- Develop computational frameworks to enable automated updating and personalising of AI algorithms by re-training from continuously recorded data from the device.
Implanted cardioverter defibrillator (ICD) devices are one of the greatest medical technological inventions of the late 20th Century. Implanted in the chest, these devices constantly monitor cardiac function, sensing lethal cardiac arrhythmias when they occur, and automatically delivering life-saving electrical shock therapy to restore normal cardiac function. Despite saving countless lives everyday from otherwise lethal cardiac episodes these devices remain far from an optimal therapy. The very high-strength electrical shocks currently required by devices drain ICD batteries. Inappropriate shocks due to mis-sensing of cardiac rhythms cause significant pain and associated psychological issues in device recipients, with frequent strong shocks leading to long-term health problems. Furthermore, the applied shock therapy often fails, and multiple therapies are required, frequently making the arrhythmia worse and leading to ultimate shock-failure and death. Thus, there is a pressing need to optimise the functioning of these devices.
Digital Twin technology, whereby in-silico replicas of patient organs are generated and used to guide, inform, monitor, diagnose and personalise medical therapies is rapidly emerging as a highly promising technology in cardiovascular medicine. ICDs represent an ideal candidate for such Digital Twin technology as they inherently continuously record patient electrical data that can be used to update and personalise computational models and optimise predictions in time as the patient’s pathology may develop. The use of Artificial Intelligence algorithms is appealing for application within ICD devices due to their ability to make fast predictions regarding sensing and decision making, utilising the wealth of continuously recorded electrical data.
In this project, we plan to construct a cohort of highly-detail patient-specific computational cardiac models. Simulation of complex arrhythmia events within these models, along with the corresponding electrical signals (as sensed by ICD devices), will be used as input to advanced AI algorithms to allow the device to learn the nature of the underlying arrhythmia and optimal electrotherapy – both application location and specific electrical stimulus protocol to deliver. These will include the latest developments in multi-stage electrotherapy, low-energy shock protocols and novel electrode designs and configurations.
Simulated pathological disease progression will then be implemented in the models to represent changes in the heart’s structure, tissue-type make-up and functional electrical remodelling, for different specific ischemic and non-ischemic cardiomyopathies. Simulated sensed electrical signals during normal (sinus) rhythm and during simulated arrhythmias will be used to develop algorithms to allow the device to learn and infer the underlying structural and functional remodelling using only the recorded electrical signals. These updated properties will then be used to re-train AI algorithms to further personalise and optimise decision-making and electrotherapy delivery. The hardware electronics requirements of ‘edge’ devices and ‘cloud’ infrastructure necessary to ‘talk’ to the ICD and facilitate re-training will be investigated to allow tangible clinical translation.
Validation of these approaches will be sought with an existing 140 patient cohort, as well as investigating opportunities to collect specific prospective data as part of similar ongoing research projects gathering similar data from these patient cohorts.
The project would suit a candidate with a keen interest in programming, mathematical/biophysical modelling and clinical application.